Drug Discovery Today d Volume 28, Number 3 d March 2023 REVIEWS

Reviews KEYNOTE REVIEW

Translating precision medicine for autism
spectrum disorder: A pressing need
Laura Pérez-Cano 1, Sara Azidane Chenlo 1,a,
Lynn Durham, MSc CEO and Founder. Lynn
Durham is a biotech entrepreneur and the

Rubén Sabido-Vera 1,a, Francesco Sirci 1, Founder and CEO of STALICLA. Driven by her
involvement with the ASD community, Lynn

Lynn Durham 1,2,⇑, Emre Guney 1,⇑ has instituted a paradigm shift in the NDD drug
discovery space by pioneering the creation of
the first NDD specific precision medicine dis-
covery platform and establishing STALICLA as
1
Discovery and Data Science (DDS) Unit, STALICLA SL, Moll de Barcelona, s/n, the first NDD focused precision medicine bio-
Edif Este, 08039 Barcelona, Spain pharmaceutical. Lynn is an inventor or co-
2
Drug Development Unit (DDU), STALICLA SA, Avenue de Sécheron 15, inventor of 11 patents and the recipient of a
brain Foundation award for “outstanding
1202 Geneva, Switzerland
leadership towards the advancement of treat-
ments for NDD patients”. She is a recognized
thought leader in the precision psychiatry
Autism spectrum disorder (ASD) is a heterogenous group of space and a reviewer for Biological psychiatry.

neurodevelopmental disorders (NDDs) with a high unmet

Emre Guney, PhD. Chief Technology Officer,
medical need. Currently, ASD is diagnosed according to Head of Discovery and Data Science Unit, DDS.
behavior-based criteria that overlook clinical and genomic Emre Guney holds 15 years of experience in
translational systems medicine, disease bioin-
heterogeneity, thus repeatedly resulting in failed clinical formatics and network pharmacology. He
spent several years as a postdoctoral scholar
trials. Here, we summarize the scientific evidence pointing at Harvard Medical School and Northeastern
to the pressing need to create a precision medicine frame- University in the USA, as well as an investi-
gator in EU funded consortia projects for drug
work for ASD and other NDDs. We discuss the role of omics toxicity and efficacy modeling during at the
Hospital del Mar Research Institute (IMIM) in
and systems biology to characterize more homogeneous Spain and at the Pharmacology & Personalised
disease subtypes with different underlying pathophysiolog- Medicine department at Maastricht University
in the Netherlands. He has a track record of
ical mechanisms and to determine corresponding tailored scientific innovation, with more than 40 peer-
reviewed publications and has authored/co-
treatments. Finally, we provide recent initiatives towards authored 4 patent applications.
tackling the complexity in NDDs for precision medicine and
cost-effective drug discovery. Laura Perez-Cano, PhD. Head of Discovery,
Deputy Head of Discovery and Data Science
Unit, DDS. Laura Perez-Cano obtained a M.S.
Keywords: Autism spectrum disorder; neurodevelopmental disorders; degree in Proteomics and a Ph.D. in Bioinfor-
omics-based data analysis; precision medicine; systems biology matics at the University of Barcelona. She
conducted her Ph.D. research work at the
Barcelona Supercomputing Center, which
resulted in important contributions to the field
of Computational Structural Biology. She
spent 5 years as a postdoctoral researcher at
the David Geffen School of Medicine at the
University of California Los Angeles, where she
co-led large-scale WGS analyses resulting in
Abbreviations: ASD, Autism spectrum disorder; NDD, Neurodevelopmental disorder; DSM-5, the identification of several novel autism risk
Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition; ADHD, Attention-deficit/ genes with an unprecedented contribution
hyperactivity disorder; LoF, Loss of function; CNV, Copy number variation; CNS, Central from inherited risk variation. She has co-
nervous system; FDR, False discovery rate; GABA, Gamma-aminobutyric acid; TD, Typically authored 4 patent applications focused on
developing; CAMP, Children’s Autism Metabolome Project; EHR, Electronic health record;
precision medicine for patients with NDDs.
DEPI, Database endophenotyping patient identification; FDA, Food and Drug Administration;
ML, Machine learning; SSRIs, Selective serotonin reuptake inhibitors; CARS, Childhood Autism
⇑ Corresponding authors. Durham, L. ([email protected]), Guney, E. ([email protected]).
a
These authors contributed equally.

1359-6446/Ó 2023 The Author(s). Published by Elsevier Ltd.

https://doi.org/10.1016/j.drudis.2023.103486 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). www.drugdiscoverytoday.com 1
 KEYNOTE (GREEN) Drug Discovery Today d Volume 28, Number 3 d March 2023

Autism spectrum disorder (ASD) is a highly heterogenous and symptoms in ASD. In this review, we summarize evidence from
lifelong neurodevelopmental disorder (NDD) that affects 1/54 existing research pointing to the clinical, genetic, molecular,
(18.5/1,000) and 1/89 (11.2/1,000) children aged around 8 years and therapeutic response heterogeneity in ASD, as well as present
in the United States and Europe, respectively, with a male to recent advances in precision medicine and discuss their potential
female ratio of 4:1 [1,2]. ASD is defined behaviorally according to enable the development of efficient drug treatments.
to the Diagnostic and Statistical Manual of Mental Disorders,
Fifth Edition (DSM-5) criteria [3] based on early onset of two core Clinical, genetic and molecular heterogeneity in ASD
Reviews KEYNOTE REVIEW

symptoms during the development age: impaired social interac- Clinical manifestation
tion and repetitive and restrictive behaviors. Although this broad ASD is a heterogeneous entity at several levels starting from the
and behaviorally-based definition is useful to stratify patients clinical point of view. It is well documented that ASD co-occurs
with ASD based on various levels of functioning it fails to frequently with other psychiatric conditions, such as attention-
acknowledge the diversity of genetic, neurophysiological, and deficit/hyperactivity disorder (ADHD, 28–31%), anxiety (20–
clinical manifestations. Moreover, it does not offer sufficient 56%), depression (11–70%), obsessive–compulsive disorder
insight into the underlying genetic or molecular causes of the (7–24%), oppositional defiant disorder (25–28%), bipolar dis-
disease, which has become a major bottleneck for the discovery order (5–7%), and schizophrenia (2–13%) [5–11]. These psy-
of consistent biomarkers and the development of efficient drug chiatric conditions co-occur in up to 70% of patients with ASD,
treatments. and have a considerable impact on the quality of life, morbidity,
Currently, the disorder remains an area of high unmet medi- and mortality [12]. Furthermore, intellectual disability is a fre-
cal need, with no treatments available to address the core symp- quent comorbidity albeit the prevalence remains controversial;
toms. To date, the only approved drugs with an ASD indication while previous epidemiological reports suggested that 70% of
(irritability associated with ASD) are aripiprazole (AbilifyÒ) and patients with ASD had co-occurring intellectual disability, due
risperidone (RisperidalÒ). However, both drugs do not address to the lack of standardized definitions and increased awareness
the core symptoms. Considering the high heterogeneity associ- and diagnoses, this number has decreased to 30% in recent
ated with ASD and multiple biological mechanisms that could reports [13]. Epilepsy also co-occurs frequently (10–26% of
potentially explain the pathophysiology of the disorder, it is patients with ASD) with a higher prevalence associated with
unlikely that any single treatment will benefit the ASD popula- older age, lower cognitive ability, poorer adaptive and language
tion as a whole. This is indeed reflected by a considerable vari- functioning, a history of developmental regression and more sev-
ability and transiency in the medications prescribed to manage ere ASD symptoms; however, no specific type of seizure shows a
the symptoms and comorbidities in patients with ASD, with significantly higher co-occurrence with ASD [14–16]. Additional
specific medications prescribed depending on coexisting comor- neurological features occurring more frequently in patients with
bidities. Therefore, precision medicine could be a promising ASD include macrocephaly, hydrocephalus, cerebral palsy,
approach for the treatment of ASD, as it offers specific drugs tai- migraine/headaches, and congenital abnormalities. In addition,
lored for specific patient subgroups to tackle different molecular 80% of patients suffer from sleep disorders and 46–84% have gas-
backgrounds. trointestinal disorders [17]. Finally, electronic health records
ASD is predominantly idiopathic or of unknown cause with (EHRs) have shown that adults with ASD are more likely to be
only 20% of the ASD population (i.e., secondary ASD) having a diagnosed with additional physical health conditions such as
specific defined cause. Remarkably, heterogeneity also applies immune conditions, obesity or stroke, compared with adults in
to secondary ASD including well-defined syndromic forms of the general population [18]. This heterogeneity associated with
ASD such as Fragile X syndrome [4]. Beyond the heterogeneity both the core and co-morbid conditions reflects a heterogeneous
observed within patients diagnosed with ASD, the current noso- pattern of neuropathology in ASD [19].
logical classification also brings a blurry definition of the bound-
aries between ASD and other psychiatric disorders. This is Genetics
exemplified by the co-occurrence of ASD and other psychiatric In line with clinical heterogeneity observed among patients diag-
disorders at rates higher than would be expected by chance. nosed with ASD, the genetic architecture of ASD is also highly
The significant degree of shared symptoms, neuropathology, complex and variable. It has been estimated that over 1,000
and genetic etiologies suggest the existence of common clusters genes might contribute to the risk of ASD [20]. This includes
of targetable biological mechanisms including different DSM-5- genes with high penetrance for ASD rarely disrupted in the gen-
defined NDDs. Taken together, due to the high heterogeneity eral population and typically resulting in monogenic ASD-related
and lack of specificity, the current behavior-based disease defini- conditions, as well as those contributing to more polygenic
tion cannot serve as a starting point for drug development. forms of ASD in which additive effects derived from common
During the past decade, different fields of medicine have variants can be determinant [21]. Over the last decades, substan-
increasingly translated omics-based findings in patients into clin- tial progress has been made towards the identification of ASD risk
ical practice. In contrast, ASD, alongside most NDDs and other genes, with most gene discovery still derived from the identifica-
neuropsychiatric conditions, is still characterized behaviorally, tion of recurrent de novo loss of function (LoF) mutations in
mostly overlooking disease etiology. Failure to acknowledge highly constrained genes, and some emerging insights from
patient’s biological heterogeneity in idiopathic ASD has repeat- the analysis of rare inherited LoF variants [22]. To date, around
edly resulted in failed clinical trials and hindered the advance- 100 genes are confidently associated (false discovery rate
ment of targeted drug development to address the core [FDR] < 0.05) with ASD risk (Supplementary Table 1) and

2 www.drugdiscoverytoday.com
 Drug Discovery Today d Volume 28, Number 3 d March 2023 KEYNOTE (GREEN)

Reviews KEYNOTE REVIEW

Drug Discovery Today

FIGURE 1
Cross-study inconsistencies across large-scale sequencing and gene expression analysis studies in ASD (A) Number of genes reported to be
significantly associated with ASD (FDR < 0.05) across one or more large-scale sequencing studies in ASD [22,92–94]. The y-axis shows the number of ASD risk
genes consistently reported in the studies listed below, with the colors of the bars representing the number of publications that have indicated those genes
as implicated in ASD risk. It can be observed from the red bar at the right of the plot, that the same 19 genes have been reported in all studies as significantly
associated with ASD, while a total of 33 genes have been identified as ASD risk genes by only one of the studies evaluated (B) Number of genes reported as
ASD differentially expressed across the studies analyzed (see Supplementary Information). Only two genes were found to be DE in all 7 studies reviewed,
whereas a total of 4572 genes were reported as DE in ASD by only one study, without being replicated in any of the 7 represented in this graph. ASD, autism
spectrum disorder; DE, differentially expressed; FDR, False Discovery Rate.

>1,000 genes (also including the former ones) show suggestive across large-scale sequencing studies in ASD (Figure 1A, Supple-
evidence from genetic and functional data (https://www.sfari.- mentary Table 1). While type I errors and the lack of statistical
com). Nevertheless, individual genes and variants associated power might explain some of these discrepancies, this is also
with elevated risk of developing ASD typically account for a very influenced by the underlying genetic heterogeneity in patients
small proportion of the cases [23]. Remarkably, genes signifi- diagnosed with ASD and the small proportion of the cases
cantly associated with ASD (FDR < 0.05) are not always replicated explained by individual genetic risk factors. Consequently, even

www.drugdiscoverytoday.com 3
 KEYNOTE (GREEN) Drug Discovery Today d Volume 28, Number 3 d March 2023

genes that are widely accepted by the scientific community to be Transcriptomics

associated with the risk of ASD (such as CUL3), do not always Over the past few years, extensive efforts have been also dedi-
reach significance in large-scale association studies. cated to the transcriptomics profiling of patients with ASD with
Functional genomic analyses have shown the convergence of the objective of detecting key dysregulated genes in both brain
clusters of ASD- and NDD-related risk genes into broad molecular and peripheral tissues. Recent studies focused on understanding
pathways which has facilitated the development of mechanistic the expression alterations involved in the pathophysiology of
hypothesis. Among those, a unifying theory proposing an neuronal cells as well as studies measuring transcriptomic

Reviews KEYNOTE REVIEW

Reviews KEYNOTE REVIEW

increased ratio of excitation versus inhibition in the brain of changes across more easily accessible tissues such as peripheral
patients with ASD was first described in 2003. This hypothesis blood for the diagnosis of ASD and the development of treat-
was supported by investigations reporting the presence of sei- ment options. As observed in genetic and metabolomic investi-
zures, unusually frequent spikes of activity, and low levels of gations, genes reported to be significantly dysregulated in ASD
gamma-aminobutyric acid (GABA) in the brain of patients diag- (FDR < 0.05) have barely been replicated across studies
nosed with ASD. However, other studies have provided contra- (Figure 1B).
dictory evidence that demonstrated an increased density of Large cohort studies that aimed at detecting a common signa-
inhibitory interneurons in post-mortem brain tissue from ture by comparing all ASD patients as a single group against a
patients with ASD, as well as hypoexcitability in Purkinje cells control population pointed to several dysregulated pathways
derived from patients with tuberous sclerosis complex and linked to the nervous system development and function, the
ASD. These contradictory results suggest that different subgroups immune system, and those related to oxidative phosphorylation
of ASD patients with an opposite pathophysiology co-exist. In and ATP synthesis [29–33]. However, similar alterations have
fact, genes tightly correlated at a molecular pathway level can been associated with other diseases, notably in patients with
result in similar phenotypical profiles predisposing to ASD, while other psychiatric conditions and cancer. Along these lines, in
mutations in distantly related ASD risk genes can result in differ- two recent studies [34,35], a strong common signal was identi-
ent or even opposite clinical manifestations (see Supplementary fied among ASD and other neurological diseases such as
Information and Supplementary Tables 2 and 3). For instance, schizophrenia and depression, linking transcriptomics changes
copy number variants (CNVs) are known to contribute to the with a shared central neuron system genetic architecture. These
heterogeneity in ASD though mirror effects that can be triggered results show the potential to detect dysregulated gene expression
by altered gene dosage, such as in 7q11.23, 22q11.21, 1q44, in patients with ASD alongside the existing convergences with
1q21.1 and 16p11.2. Overall, accumulating evidence suggests gene expression dysregulations reported in other neurological-
the existence of subgroups of ASD patients with different or even and immune-related diseases. They also arguably point to the
opposite pathophysiology, underlining the need to account for lack of disease-specific information to drive precise treatment
the type and directionality of the perturbations across patients development in ASD.
for the development of efficient and safe drug treatments in ASD.

Need for patient stratification and clinical endpoint

Metabolomics refinement for drug discovery
Metabolomics, the profiling of small-molecule metabolites, has The complexity at the clinical, genetic, and molecular level
emerged as a major contributor in the study of ASD pathogene- underlying ASD poses a significant challenge for drug discovery.
sis. Indeed, the metabolome reflects the previously discussed This is reflected by the lack of success in identifying treatments
interaction between genetic [24] and environmental influences that address the core symptoms of ASD (Supplementary Table 4,
[25] in ASD, owing to its sensitivity to capture interactions adapted from Siafis et al. [36]).
between the genome, gut microbiome, diet, and environmental One of the main difficulties in ASD and NDD drug develop-
factors, and therefore can provide information to bridge the ment is the complexity of genetic and environmental factors
gap between genotype and phenotype [26]. In this context, inducing early developmental alterations in the brain which
promising ASD diagnostic biomarkers (such as metabolites) can involve dysregulation and imbalance of several central ner-
could play a critical role in earlier disease diagnosis during vous system (CNS) neurotransmitters such as serotonin, acetyl-
infancy, enabling the initiation of early intervention which has choline, histamine, dopamine, GABA, and glutamate [37].
been shown to promote improvements in the behavioral symp- Abnormalities in prefrontal cortex and mesolimbic circuit
tomatology of children with ASD [27]. involving dopamine as a key mediator have an impact on behav-
Recently, the study of differential metabolites between ior and emotional regulation, while alterations in other relevant
patients with NDDs and typically developing (TD) individuals neurotransmission systems can result in reduced GABAergic gene
has attracted growing attention. This strategy has been widely expression and/or an increase in glutaminergic signaling. The
used in the analysis of different samples (such as blood, urine, unpaired modulation of these different brain circuits in ASD
and stools) from patients with ASD and healthy subjects [28]. leads to the development of core symptoms and a spectrum of
However, the findings across various studies analyzing changes several associated comorbidities like depression, schizophrenia,
in blood metabolites in children with ASD have been rarely con- seizures, insomnia, impaired behavioral activity, attention deficit
sistent (see Supplementary Information). This lack of repro- and self-injury. Pharmacological targeting of the ASD markers is
ducibility across studies is in part due to the high currently applied in several clinical studies with potential thera-
heterogeneity among patients diagnosed with ASD. peutic strategies [37–39]. Although approved treatments can

4 www.drugdiscoverytoday.com
 Drug Discovery Today d Volume 28, Number 3 d March 2023 KEYNOTE (GREEN)

help ameliorate related symptoms and comorbidities, these treat- inconclusive results of a randomized, double-blind, placebo-
ments do not address the core symptoms and may cause several controlled, phase-2 clinical study of arbaclofen administrated
off-target and side effects. In this regard, risperidone (RisperdalÒ) to patients with ASD who had a history of seizure disorder [47].
and aripiprazole (AbilifyÒ) are the only drugs approved by the The authors highlighted the importance of narrowing the target
Food and Drug Administration (FDA) for patients with ASD. population to detect the improvements only expected in a por-
These medicines can be prescribed to mitigate irritability that tion of the heterogeneous autism spectrum. Post hoc analysis
can be associated with ASD. Other commonly prescribed medici- on socialization score focusing only on consistent raters of Vine-

Reviews KEYNOTE REVIEW

nes include selective serotonin reuptake inhibitors (SSRIs), anti- land Adaptive Behavior Scale-II scores demonstrated significant
anxiety medications or stimulants, although none have received improvement under arbaclofen administration after 12 weeks.
FDA approval for the treatment of ASD. A similar attempt was made by repurposing a drug used for the
Recent studies suggest that only a subset of the patients will treatment of hypertension known as bumetanide, a brain Na-K-
respond to any given treatment and the chance that one or Cl cotransporter (NKCC1) and renal NKCC2 chloride-importers
few therapeutic treatments will be effective on the entire ASD inhibitor that reduces intracellular chloride concentration and
population is very narrow [40]. Thus, the paradigm in ASD promotes GABAergic inhibition. The efficacy of bumetanide
should move from a “one-size-fits-all” approach towards a more was assessed in respective clinical studies involving children with
personalized medicine approach, in which the identification of ASD or Asperger syndrome. These studies demonstrated signifi-
multiple “biologies” underlying ASD will be key to identify cant improvement in Childhood Autism Rating Scale (CARS),
molecular targets involved in the pathophysiology of specific Clinical Global Impression (CGI) improvement and Social
patient subgroups with the potential to respond to a particular Responsiveness Scale (SRS) scores compared with placebo [49].
treatment based on a shared biology. In addition, data related However, response based on comorbidities could not be analyzed
to the failure of clinical trials in ASD can be associated with sub- due to the rather small sample set (n < 100) [49]. More recently,
optimal study design and outcome measures derived from broad the largest European phase 3 clinical trial of bumetanide for the
diagnostic criteria rather than clinically meaningful or patient treatment of core symptoms of ASD in pediatric population was
centric criteria. Historically, clinical trials testing for drug candi- terminated due to lack of efficacy. Servier, the sponsor of the
dates in ASD have included broad groups of participants result- study, reported in a press release [50] that the results did not
ing in failure to demonstrate clinical benefit among all study show efficacy of bumetanide compared with placebo for the
participants [40,41]. Similarly, existing clinical studies in ASD treatment of the general ASD population. The main criteria for
typically focus on symptoms observed in ASD patients such as assessing core symptoms of ASD were alterations in communica-
irritability, hyperactivity, and anxiety as clinical endpoints, tion and social interactions with the presence of stereotypical
rather than molecular mechanisms underlying these symptoms and repetitive behaviors. Prof. Ben-Ari, who is the president of
[42]. More recently, the core symptoms of ASD such as social Neurochlore, the company that originally partnered with Servier
withdrawal and social communication are increasingly being on the bumetanide clinical trial, stated that “the heterogeneity of
considered [42]. However, despite these recent attempts to target ASDs probably makes it impossible to offer a sole treatment for
the core deficits, there is still no substantial evidence supporting all autistic children”.
the efficacy of drug treatments addressing these core symptoms Taken together, these studies suggest the need for novel data-
at the population-level. driven approaches for the identification of sub-populations of
In this context, in clinical studies evaluating D-cycloserine patients diagnosed with ASD and characterization of more robust
[43] and memantine [44] did report clinically meaningful effects endpoints that account for the molecular alterations underlying
in reducing core symptoms in some adolescents and adults with these sub-populations. Accordingly, we expect to see an increas-
ASD. SSRIs are also often prescribed to patients with Fragile X ing role for artificial intelligence based analyzes of social func-
syndrome to relieve anxiety, irritability, improve mood, and tioning, repetitive/restricted behavior, and cognitive
social deficits. Polymorphism of five genes involved in the sero- impairment scores combined with multi-omics data. In parallel,
tonin pathway and its metabolism have been studied and signif- clinical studies will need to employ more objective and measur-
icantly correlated with improved cognitive and behavioral scores able endpoints to counteract the high placebo effect. Endpoints
in young children with Fragile X syndrome treated with sertra- derived from quantitative measurements such as EEG signals,
line (compared with placebo) were associated with specific BNDF levels of biomolecules in whole blood and cererbospinal fluid,
genotypes [45]. Overall, the variation observed among patients and from longitudinal measures assessing cognitive, emotional,
in their behavioral response to treatments points to the necessity motor and sensory functions screened by the NIH toolbox plat-
to fully consider the heterogeneity in the etiology, pathogenesis, form (https://neuroscienceblueprint.nih.gov/resources-tools/
and symptomatology of ASD patients to successfully advance blueprint-resources-tools-library/nih-toolbox-assessment-neuro-
therapeutics [46]. logical-and) will play a critical role in informing on the biological
Heterogeneity in terms of treatment response was also under- mechanisms underlying specific behavioral patterns.
lined by drug treatments targeting the proposed excitation versus
inhibition imbalance in ASD, as reported in the arbaclofen and
bumetanide trials [47–49]. Arbaclofen, a GABA-B receptor ago- Computational models for characterizing patients
nist, was tested for the treatment of Fragile X syndrome based with ASD and developing therapeutic interventions
on previous evidence of behavior improvement in genetic stud- Computational methods can help improve the characterization
ies in mouse models [43]. In 2017, VanderWeele et al. reported and classification of patients diagnosed with ASD. Stratification

www.drugdiscoverytoday.com 5
 KEYNOTE (GREEN) Drug Discovery Today d Volume 28, Number 3 d March 2023
Reviews KEYNOTE REVIEW

Drug Discovery Today

FIGURE 2
Schematic representation of precision medicine in NDDs using systems biology and computational data analysis. Multi-omics and clinical data are
integrated and enriched using systems biology based multimodal data analysis. Machine learning models including both unsupervised & supervised models
are used to identify the genetic and clinical characteristics underlying specific patient subgroups leading to patient stratification, biomarker, and drug
discovery. NDD, neurodevelopmental disorders.

of patients into subgroups of interest offers a promising strategy teachers whenever applicable). These evaluations consider a
to tackle the inherent heterogeneity in ASD and to establish child's current behavior and developmental history to measure
more homogeneous disease subtypes; nevertheless, the most fre- social and communication abilities. The data collected through
quently used variables for subtyping analyses conducted so far such behavioral and observational assessments in the form of
are based on behavioral assessments [51]. In this regard, discov- questionnaires or list of clinical criteria is often categorical in nat-
ery of the patient subgroups with common molecular back- ure. Moreover, the collected information does not offer insight
grounds would be key to develop targeted and more effective into the genetic and molecular disruptions underlying the
therapies. Over the past decade, multiple approaches have been behavioral and observational characteristics, failing to assess
proposed, which include both supervised learning approaches the quantitative features relevant to the disease condition
that involve training models for the classification of patients appropriately.
with ASD into certain categories and unsupervised learning In this context, one of the first studies using a machine learn-
approaches that cluster patients with ASD into more homoge- ing (ML) classifier on ADOS generic data concluded that 8 of the
neously defined subgroups (Figure 2). This has given rise to the 29 items of the questionnaire were sufficient to classify autism
emergence of computational psychiatry, aiming to improve diag- patients with high sensitivity and specificity values, which can
nosis through computational analysis of large-scale behavioral help shorten the time for diagnosis [53] and reduce the time
and biological data sets (see a comprehensive review by Jacob for clinical assessment. A recent literature review of ML
and colleagues) [52]. The main limitation with these approaches, approaches for ASD classification using behavioral data (22 stud-
however, is that the computational models rely on small sample ies assessed) highlighted that even if very good metrics were
sizes across patient datasets, particularly given the biological and achieved by ML models, no evidence supports their readiness
clinical heterogeneity. for clinical use [54]. Overall, the validity and generalizability of
Given the challenges associated with the diagnosis of ASD these models based on behavioral data from small cohorts war-
using behavioral and observational assessment, the clinical diag- rant further research. On the other hand, a study suggested that
nostic data for ASD relies on various standardized psychological computational methods based on more quantitative and biolog-
assessment tools such as the Autism Diagnostic Interview- ically informative features such as pregnancy follow-up ultra-
Revised, the Autism Diagnostic Observation Schedule (ADOS), sound and biological measurements might provide an early
SRS and CARS. Children with a suspected diagnosis of ASD are prognosis of ASD to enable early behavioral interventions that
usually evaluated within a developmental framework, including can efficiently attenuate ASD developmental sequels [55]. Never-
multiple informants from diverse contexts (e.g., parents and theless, problems such as data repository discrepancies, limita-

6 www.drugdiscoverytoday.com
 Drug Discovery Today d Volume 28, Number 3 d March 2023 KEYNOTE (GREEN)

tions inherent to clinical assessment instruments that treat ASD cal, clinical, and demographic datasets [69]. Often, however,
as a general diagnostic category, and overfitting to the training deep learning-based methods require large sample sizes and lack
data remain as potential challenges to improve these models. mechanistic interpretation in regard to the applicability in a clin-
The clinical assessment tools used for ASD categorization in ical setting. Taken together with the high heterogeneity in ASD,
ML models have also been used in neuroimaging data analysis, further research will be required to validate previously character-
given accumulating evidence that suggests the potential utility ized ASD subgroups and the associated therapeutic implications.
of neuroimaging in capturing atypical brain organization in Nevertheless, the integration and use of different computational

Reviews KEYNOTE REVIEW

ASD. These variations can occur either at the structural level such methods will provide valuable opportunities to extend the
as variations in cortical thickness, surface area or brain volume in domain of applicability of each method and more thoroughly
patients with CNV in certain ASD risk regions including 7q11.23, exploit information coming from diverse sources.
22q11.21, 1q44, 1q21.1 and 16p11.2 [56–58] or in pattern con-
nectivity abnormalities revealed by functional neuroimaging
techniques. A recent study using functional magnetic resonance Recent omics- and endophenotyping-based methods
imaging (MRI) on children aged 6 months with a high familial towards precision medicine
risk for ASD, was able to predict diagnostic outcomes at age 2 Omics-based methods are increasingly gaining popularity to pro-
(specificity: 100%, sensitivity: 81.8%) [59]. In a parallel study, a cess large cohort datasets towards precision medicine, driven by
deep-learning algorithm on MRI data from high-risk children the success of their application on patients with cancer, both for
aged 6–12 months was able to predict autism onset at 24 months disease subtyping and for determining the most appropriate
(specificity: 95%, sensitivity: 88%) [60]. Provided that several dif- treatments [70–72]. More and more initiatives across a wide
ficulties such as reduced sample sizes, standardization of imaging number of different conditions [73–76] have recently embarked
methods or cross-center validation issues are solved, further upon driving the translational application of precision medicine
replication of these studies in the future could result in the devel- with special emphasis on integrating omics-based data with clin-
opment of reliable imaging biomarkers. Interestingly, two studies ical information from EHRs and deep phenotyping of patients.
have tried to use neuroimaging data to discern ADHD from ASD These studies support existence of endophenotypes (i.e., inter-
owing to the clinical cooccurrence and genetic overlap between mediate phenotypes corresponding to biological processes)
these 2 conditions [12,61,62]. This has motivated the emergence linked to genetic and molecular changes underlying the disease
of studies aimed to identify homogeneous ASD subgroups based pathology [77], beyond neuropsychiatric conditions. More
on neuroimaging features, also known as neurosubtyping. importantly, the endophenotypes demonstrate how data inte-
Results from initial efforts do support the feasibility and poten- gration and processing can detect recurrent patient subgroups,
tial utility of neurosubtyping, even though there are still some to enable better diagnosis and personalized treatment, also for
technical limitations [63,64]. NDDs.
Heterogeneity remains a fundamental issue for the character- The application of endophenotyping towards identification of
ization of patients with ASD as some biological markers may only better treatment alternatives has attracted interest in the past few
be found within a subset of patients, even though there are years [73]. Using network medicine tools [78–80], numerous
clinical-level commonalities manifesting across patient sub- studies have investigated repositioning of drugs for cardiovascu-
groups. Accordingly, several reports document stratification of lar, cerebrovascular, and other CNS disorders [81–83]. In fact, of
patients into subgroups using computation techniques applied all medical specialties, psychiatry has arguably the most to gain
on the clinical data. For example, Feczko et al. [65] attempted by incorporating mechanistic knowledge on endophenotypes
to identify and characterize cognitive subtypes within the ASD to identify precision medicine treatments, considering substan-
population using a Functional Random Forest ML classification tial heritability of psychiatric phenotypes and the current reli-
model, which revealed 3 ASD and 4 TD putative subgroups with ance on syndrome-based diagnoses. While there are certainly
distinct behavioral profiles in a cohort of 47 children diagnosed challenges to overcome, omics-driven analyses provide an empir-
with ASD and 58 TD children. In another study, Narita et al. [66] ical framework upon which psychiatry can now progress towards
conducted cluster-based genome wide association studies on 712 better understanding of the disease mechanisms, better treat-
probands and 354 controls and identified 65 chromosomal loci ments, and to better ways of targeting treatments to the patients
of interest, with some of these loci being located within or near most likely to benefit from, thus paving the way for precision
previously reported candidate genes for ASD. These findings sug- psychiatry [84]. Accordingly, efforts to integrate omics and clin-
gest that clustering may successfully identify subgroups with ical data towards precision medicine are increasingly being
homogeneous disease etiologies. Furthermore, Luo et al. [67] applied to have a detailed view on gene-phenotype relationships.
were able to find a subgroup of the patients with dyslipidemia- For instance, Karczewski et al. integrated whole exome sequenc-
associated autism using a multidimensional precision medicine ing data from >200,000 individuals with deep phenotyping data
approach, where healthcare claims, EHRs, familial whole- to analyze a single mutation effect on >3,000 annotated pheno-
exome sequences, and neurodevelopmental gene expression pat- types, providing the scientific community with an unprece-
terns were combined. Other applications are emerging in the dented view of gene-phenotype relationships [85]. These efforts
past few years, such as those based on deep learning [68], which to identify and quantify the relationship between genetic vari-
can be widely used to predict personalized drug response and ants and patients’ phenotypes are currently delivering promising
optimize medication selection and dosing using knowledge results in rare diseases, enabling prompt diagnosis and personal-
extracted from the large and complex molecular, epidemiologi- ized decisions about the treatment for patients. A representative

www.drugdiscoverytoday.com 7
 KEYNOTE (GREEN) Drug Discovery Today d Volume 28, Number 3 d March 2023

example is the study by Sweeney et al., where they showed that [88–89]. A previous study on the Children’s Autism Metabolome
the surgical operation could have been determined by a prompt Project (CAMP) cohort was able to define an altered metabolic
application of a rapid diagnostic protocol based on whole gen- phenotype consisting of imbalanced branch chain amino acids
ome sequencing data [86]. metabolism in 16.7% of the CAMP participants, with specificity
A deep understanding of omics-based molecular phenotypes and precision values of 96.3% and 93.5%, respectively [87]. Even
could also provide a portfolio of biomarkers suitable for drug though additional research is needed, it is not clear, for example,
development and clinical trial approaches in NDDs. Omics has how stable these metabotypes are in a particular child [90].
Reviews KEYNOTE REVIEW

opened the door to a vast amount of information about func- Nonetheless, metabolomics data analysis has shown an immense
tion, protein and genetic interactions, gene product expression, potential in early diagnosis of ASD and further research can have
metabolite and lipid content, and complex feedback processes a profound impact on patients’ treatment and prognosis [91].
that integrate these molecules into pathways and in time and Latest research has also tackled the transcriptomic profiling of
space. Understanding the entire picture will allow us to design individuals with ASD from a precision medicine perspective. To
and test molecular biomarkers for response to different therapeu- this end, dissecting ASD into different subtypes and characteriz-
tic strategies and may allow the development of personalized ing corresponding subtype-specific omics-based signatures to
medicine strategies to contribute to the success of clinical trials explain the phenotypic variability in these subtypes constitute
in ASD. a promising strategy for the diagnosis and personalized
Along these lines, metabotyping, i.e., subtyping based on treatment.
shared metabolic phenotypes identified using metabolic Progress made in the omics and computational biology fields
biomarkers associated with the risk for ASD, can enable stratifica- has also enabled emergence of not only several pioneering con-
tion of the disorder into distinct subpopulations based on a com- sortia projects towards characterizing genetic factors underlying
mon metabolic dysregulation [87] and lead to potential ASD, but also various clinical stage biotech companies focused
therapeutic targets for individuals with a specific metabotype on the development of precision medicine-based diagnoses and

Drug Discovery Today

FIGURE 3
DEPI platform overview. Developed by STALICLA, Databased Endophenotyping Patient Identification (DEPI) allows the stratification and identification of
subgroups of patients with NDDs, by integrating different analysis approaches such as system biology, EHR and multi-omics profiling, and using patient data
such as biosamples and clinical questionnaires (patient profiling). On the other hand, drug screening in specific cell lines (drug profiling) can address the
therapeutic needs of a specific group, thus generating tailored treatments for each subgroup. EHR, electronic health records; NDD, neurodevelopmental
disorders.

8 www.drugdiscoverytoday.com
 Drug Discovery Today d Volume 28, Number 3 d March 2023 KEYNOTE (GREEN)

treatment options for patients with ASD and other NDDs. For various studies in which patients with ASD were treated indis-
example, public initiatives such as SFARI (https://www.sfari. tinctively despite clear biological and clinical differences. The
org), iHART (http://ihart.org) or MSSNG (https://research.mss. poor overlap between the genetic- and molecular-level perturba-
ng) have collected medical and genetic information from tions across different cohorts reflects the limitations of existing
>10,000 patients with ASD. On the other hand, companies clinical studies that use ASD as a broad diagnostic category and
such as Stemina Biomarker Discovery (https://stemina.com/), emphasizes the urgent need to stratify the heterogenous ASD
integrates the NeuroPointDX technology, a metabolomics- population into biologically-defined ASD subgroups to improve

Reviews KEYNOTE REVIEW

based ASD diagnostic tool (NCT02548442); quadrant biosciences the clinical response success rate. The recent paradigm shifts
with Clarifi ASDÒ, also develops an ASD diagnostic tool toward “multi-omics” data integration together with systematic
based on epigenomics (https://quadrantbiosciences.com/; and systems biology based analyses of large-scale data can facili-
NCT02832557); and STALICLA (https://stalicla.com/) offers tate identification of ASD patient subgroups with a similar under-
Database Endophenotyping Patient Identification (DEPI), a sys- lying “biology”, which may allow the identification of consistent
tems biology and multi-omics-based platform that allows stratifi- biomarkers. Such data-driven analyses can be used to translate
cation of NDD patients into biologically similar and clinically mechanistic perturbations underlying different patient sub-
actionable patient subgroups (i.e., patients that share similar groups into actionable molecular targeting strategies. Therefore,
pathway-level alterations) and the exploration of the corre- to advance the identification of novel personalized treatments
sponding tailored therapeutic treatments for these patients in and overcome the current challenges and pitfalls experienced
each subgroup (Figure 3). in past ASD clinical studies, it is critical to apply robust modeling
One of the first applications of DEPI technology is STP1, the strategies that account for heterogeneity by ensuring cross-
first precision medicine therapeutic package in the ASD space cohort replication and by using interpretable models that are
specifically developed for the treatment of patients matching not prone to overfitting, especially for small cohorts. Similarly,
the criteria for a clinically and biologically validated ASD sub- there is a need for quantifiable biomarkers as primary endpoints
group, for which a Phase1b interventional clinical trial has to characterize mechanistically changes in core deficits upon
recently been completed with success (NCT04644003). DEPI treatment across patient subgroups. Quantifiable endpoints will
leverages proprietary NDD-tailored knowledge bases and combi- allow to assess the effect of treatments more accurately and cir-
nes clinical and biomedical data from large-scale NDD cohorts cumvent the high placebo effect observed in some past clinical
for endophenotyping, clustering and drug positioning (Figure 3). trials.
The endophenotypes identified by DEPI, also named actionable Along these lines, recent studies and public and private trans-
clinical signs and symptoms, constitute quantitative biological lational initiatives open a new era in the diagnosis and treatment
traits with a strong genetic component that significantly con- of ASD and other NDDs, offering solutions to fill the existing
tributes to the risk of NDDs. These are based on curated and translational gap to improve the quality of life of the patients
NDD-tailored catalogs that combine clinical, genetic, molecular, with more severe forms of these conditions.
and pathway-level data. The clinical data include information
extracted from EHRs such as medical history, where Data availability
behaviorally-defined conditions are excluded using semantic- The references from which the data is collected for the supple-
type matching based on standardized medical ontologies. Impor- mentary tables are listed in the supplementary information.
tantly, DEPI can further integrate patient-derived biomarkers
with cell-based multi-omics data to prioritize drug treatments
Acknowledgments
and enable the matching of the right patients with a NDD with
The authors wish to thank Edgard Verdura, Mattia Bosio, and
the right drug treatments. The broad applicability of DEPI is
Jean-Marc Hyvelin, discussions with whom have contributed
being investigated in an ongoing observational study consisting
shaping the manuscript. The authors are grateful to Anezka Zaji-
of 250 patients (NCT04273087) for STP1 and STP2, the two pre-
cova for her support on graphical design.
cision medicine packages targeting two ASD subpopulations.

Conclusion Conflict of interests

The clinical, genetic, and molecular heterogeneity observed in All authors are employees of STALICLA.
patients diagnosed with ASD and the application of a “one-
size-fits-all” approach in the past clinical trials have hindered Appendix A. Supplementary material
the development of efficient drugs for the treatment of patients Supplementary material to this article can be found online at
with ASD and other NDDs. In this review, we have highlighted https://doi.org/10.1016/j.drudis.2023.103486.

References
[1] Maenner, M.J. et al. (2020) Prevalence of autism spectrum disorder among Posada_Introduction-to-the-ASDEU-programme-and-preliminary-results-of-
children aged 8 years—autism and developmental disabilities monitoring studies-into-prevalence-and-cost.pdf. Published September 25, 2018. Accessed
network, 11 sites, United States, 2016. MMWR Surveill. Summaries 69, 1. November 30, 2022.
[2] An EU Strategy for Autism to leave No One Behind Introduction to the ASDEU [3] Association, A.P. (2013) Diagnostic and statistical manual of mental disorders:
programme and preliminary results of studies into prevalence and cost (2018) DSM-5. Arlington, VA. 2013.
Website: https://www.autismeurope.org/wp-content/uploads/2018/09/M_

www.drugdiscoverytoday.com 9
 KEYNOTE (GREEN) Drug Discovery Today d Volume 28, Number 3 d March 2023

[4] Verdura, E. et al. (2021) Heterogeneity in Fragile X Syndrome highlights the [35] Guan, J. et al. (2019) Commonality in dysregulated expression of gene sets in
need for precision medicine-based treatments. Front. Psychiatry 1661. cortical brains of individuals with autism, schizophrenia, and bipolar disorder.
[5] Gargaro, B.A. et al. (2011) Autism and ADHD: how far have we come in the Translat. Psychiatry 9, 1–15.
comorbidity debate? Neurosc. Biobehav. Rev. 35, 1081–1088. [36] Siafis, S. et al. (2022) Pharmacological and dietary-supplement treatments for
[6] Leitner, Y. (2014) The co-occurrence of autism and attention deficit autism spectrum disorder: a systematic review and network meta-analysis. Mol.
hyperactivity disorder in children–what do we know? Front. Hum. Neurosci. 8, Autism 13, 1–17.
268. [37] Eissa, N. et al. (2018) Current enlightenment about etiology and
[7] Rosen, T.E. et al. (2018) Co-occurring psychiatric conditions in autism spectrum pharmacological treatment of autism spectrum disorder. Front. Neurosci. 12,
disorder. Int. Rev. Psychiatry 30, 40–61. 304.
Reviews KEYNOTE REVIEW

[8] Simonoff, E. et al. (2008) Psychiatric disorders in children with autism spectrum [38] Alfageh, B.H. et al. (2019) Safety and tolerability of antipsychotic medication in
disorders: prevalence, comorbidity, and associated factors in a population- individuals with autism Spectrum disorder: a systematic review and meta-
derived sample. J. Am. Acad. Child Adolescent Psychiatry 47, 921–929. analysis. Pediatric Drugs 21, 153–167.
[9] Joshi, G. et al. (2010) The heavy burden of psychiatric comorbidity in youth [39] Turner, M. (2020) The role of drugs in the treatment of autism. Austr. Prescriber
with autism spectrum disorders: A large comparative study of a psychiatrically 43, 185.
referred population. J. Autism Dev. Disorders 40, 1361–1370. [40] Beversdorf, D.Q. (2016) CONSORTIUM MAS. Phenotyping, etiological factors,
[10] Kirsch, A.C. et al. (2020) Association of comorbid mood and anxiety disorders and biomarkers: toward precision medicine in autism spectrum disorders. J.
with autism spectrum disorder. J. Am. Med. Assoc. Pediatrics 174, 63–70. Dev. Behav. Pediatrics 37, 659.
[11] Lai, M.-C. et al. (2014) Autism. The Lancet 383, 896–910. DOI: 10.1016/S0140- [41] The Flawed Designs of Drug Trials for Autism (2017) Website: https://www.
6736(13)61539-1. theatlantic.com/health/archive/2017/02/autism-drugs/516855/. Published
[12] Lai, M.-C. et al. (2019) Prevalence of co-occurring mental health diagnoses in February 16, 2017. Accessed November 16, 2021.
the autism population: a systematic review and meta-analysis. The Lancet [42] McCracken, J.T. et al. (2021) Drug development for autism spectrum disorder
Psychiatry 6, 819–829. DOI: 10.1016/S2215-0366(19)30289-5. (ASD): progress, challenges, and future directions. Eur. Neuropsychopharmacol.
[13] Baio, J. et al. (2018) Prevalence of autism spectrum disorder among children 48, 3–31.
aged 8 years—autism and developmental disabilities monitoring network, 11 [43] Schade, S. and Paulus, W. (2016) D-cycloserine in neuropsychiatric diseases: a
sites, United States, 2014. MMWR Surveill. Summaries 67, 1. systematic review. Int. J. Neuropsychopharmacol. 19.
[14] Ewen, J.B. et al. (2019) Epilepsy and autism severity: a study of 6,975 children. [44] Aman, M.G. et al. (2017) Safety and efficacy of memantine in children with
Autism Res. 12, 1251–1259. autism: randomized, placebo-controlled study and open-label extension. J.
[15] Lukmanji, S. et al. (2019) The co-occurrence of epilepsy and autism: A Child Adolescent Psychopharmacol. 27, 403–412.
systematic review. Epilepsy Behav. 98, 238–248. [45] AlOlaby, R.R. et al. (2017) Molecular biomarkers predictive of sertraline
[16] Viscidi, E.W. et al. (2013) Clinical characteristics of children with autism treatment response in young children with fragile X syndrome. Brain Dev. 39,
spectrum disorder and co-occurring epilepsy. PLoS One 8, e67797. 483–492.
[17] Al-Beltagi, M. (2021) Autism medical comorbidities. WORLD J. Clin. Pediatrics [46] Singh, K. et al. (2014) Sulforaphane treatment of autism spectrum disorder
10, 15. (ASD). Proc. Natl. Acad. Sci. 111, 15550–15555.
[18] Croen, L.A. et al. (2015) The health status of adults on the autism spectrum. [47] Veenstra-VanderWeele, J. et al. (2017) Arbaclofen in children and adolescents
Autism 19, 814–823. with autism spectrum disorder: a randomized, controlled, phase 2 trial.
[19] Amaral, D.G. et al. (2008) Neuroanatomy of autism. Trends Neurosci. 31, Neuropsychopharmacology 42, 1390–1398.
137–145. [48] Lemonnier, E. et al. (2012) A randomised controlled trial of bumetanide in the
[20] He, X. et al. (2013) Integrated model of de novo and inherited genetic variants treatment of autism in children. Translat. Psychiatry 2, e202.
yields greater power to identify risk genes. PLoS Genet. 9, e1003671. [49] Lemonnier, E. et al. (2017) Effects of bumetanide on neurobehavioral function
[21] Weiner, D.J. et al. (2017) Polygenic transmission disequilibrium confirms that in children and adolescents with autism spectrum disorders. Translat. Psychiatry
common and rare variation act additively to create risk for autism spectrum 7, e1056.
disorders. Nature Genet. 49, 978–985. [50] Servier and Neurochlore announce the main results of the two phase 3 clinical
[22] Sanders, S.J. et al. (2015) Insights into autism spectrum disorder genomic studies assessing bumetanide in the treatment of Autism Spectrum Disorders in
architecture and biology from 71 risk loci. Neuron 87, 1215–1233. children and adolescents (2021) Website: https://mma.prnewswire.com/media/
[23] de la Torre-Ubieta, L. et al. (2016) Advancing the understanding of autism 1609162/PR_Servier_Neurochlore__bum_tanide_Phase_3_Results.pdf.
disease mechanisms through genetics. Nature Med. 22, 345–361. Published September 7, 2021. Accessed November 28, 2022.
[24] Geschwind, D.H. and State, M.W. (2015) Gene hunting in autism spectrum [51] van Rentergem, J.A.A. et al. (2021) Validation strategies for subtypes in
disorder: on the path to precision medicine. The Lancet Neurology 14, psychiatry: A systematic review of research on autism spectrum disorder. Clin.
1109–1120. Psychol. Rev. 102033.
[25] Sharon, G. et al. (2019) Human gut microbiota from autism spectrum disorder [52] Jacob, S. et al. (2019) Neurodevelopmental heterogeneity and computational
promote behavioral symptoms in mice. Cell 177, 1600–1618.e17. approaches for understanding autism. Translat. Psychiatry 9, 1–12.
[26] Orozco, J.S. et al. (2019) Metabolomics analysis of children with autism, [53] Wall, D.P. et al. (2012) Use of machine learning to shorten observation-based
idiopathic-developmental delays, and Down syndrome. Translat. Psychiatry 9, screening and diagnosis of autism. Translat. Psychiatry 2, e100.
1–15. [54] Cavus, N. et al. (2021) A systematic literature review on the application of
[27] Warren, Z. et al. (2011) A systematic review of early intensive intervention for machine-learning models in behavioral assessment of autism spectrum
autism spectrum disorders. Pediatrics 127, e1303–e1311. disorder. J. Personal. Med. 11, 299.
[28] Shen, L. et al. (2020) Biomarkers in autism spectrum disorders: Current progress. [55] Caly, H. et al. (2021) Machine learning analysis of pregnancy data enables early
Clin. Chim. Acta 502, 41–54. identification of a subpopulation of newborns with ASD. Sci. Rep. 11, 1–14.
[29] Chow, M.L. et al. (2012) Age-dependent brain gene expression and copy [56] Malhotra, D. and Sebat, J. (2012) CNVs: harbingers of a rare variant revolution
number anomalies in autism suggest distinct pathological processes at young in psychiatric genetics. Cell 148, 1223–1241.
versus mature ages. PLoS Genet. 8, e1002592. [57] Sonderby, I.E. et al. (2020) Dose response of the 16p11. 2 distal copy number
[30] Garbett, K. et al. (2008) Immune transcriptome alterations in the temporal variant on intracranial volume and basal ganglia (vol 25, pg 584, 2018). Mol.
cortex of subjects with autism. Neurobiol. Dis. 30, 303–311. Psychiatry 25, 692–695.
[31] Ivanov, H.Y. et al. (2015) Blood-based gene expression in children with autism [58] Maillard, A. et al. (2015) The 16p11. 2 locus modulates brain structures common
spectrum disorder. Biodiscovery 17, e8966. to autism, schizophrenia and obesity. Mol. Psychiatry 20, 140–147.
[32] Liu, X. et al. (2016) Disruption of an evolutionarily novel synaptic expression [59] Emerson, R.W. et al. (2017) Functional neuroimaging of high-risk 6-month-old
pattern in autism. PLoS Biol. 14, e1002558. infants predicts a diagnosis of autism at 24 months of age. Sci. Translat. Med. 9.
[33] Ramaswami, G. et al. (2020) Integrative genomics identifies a convergent [60] Hazlett, H.C. et al. (2017) Early brain development in infants at high risk for
molecular subtype that links epigenomic with transcriptomic differences in autism spectrum disorder. Nature 542, 348–351.
autism. Nature Commun. 11, 1–14. [61] Lim, L. et al. (2013) Disorder-specific predictive classification of adolescents
[34] Gandal, M.J. et al. (2018) Shared molecular neuropathology across major with attention deficit hyperactivity disorder (ADHD) relative to autism using
psychiatric disorders parallels polygenic overlap. Science 359, 693–697. structural magnetic resonance imaging. PLoS One 8, e63660.

10 www.drugdiscoverytoday.com
 Drug Discovery Today d Volume 28, Number 3 d March 2023 KEYNOTE (GREEN)

[62] Ghiassian, S. et al. (2016) Using functional or structural magnetic resonance [80] Aguirre-Plans, J. et al. (2019) GUILDify v2. 0: a tool to identify molecular
images and personal characteristic data to identify ADHD and autism. PLoS One networks underlying human diseases, their comorbidities and their druggable
11, e0166934. targets. J. Mol. Biol. 431, 2477–2484.
[63] Hong, S.-J. et al. (2020) Toward neurosubtypes in autism. Biol. Psychiatry 88, [81] Fang, J. et al. (2020) Harnessing endophenotypes and network medicine for
111–128. Alzheimer's drug repurposing. Med. Res. Rev. 40, 2386–2426.
[64] Pagani, M. et al. (2021) mTOR-related synaptic pathology causes autism [82] Fang, J. et al. (2021) Endophenotype-based in silico network medicine discovery
spectrum disorder-associated functional hyperconnectivity. Nature Commun. combined with insurance record data mining identifies sildenafil as a candidate
12, 1–15. drug for Alzheimer’s disease. Nature Aging 1, 1175–1188.
[65] Feczko, E. et al. (2018) Subtyping cognitive profiles in autism spectrum disorder [83] Kalueff, A.V. et al. (2014) Rethinking CNS disorders: time for new drug targets.

Reviews KEYNOTE REVIEW

using a functional random forest algorithm. Neuroimage 172, 674–688. Trends Pharmacol. Sci. 35, 491–492.
[66] Narita, A. et al. (2020) Clustering by phenotype and genome-wide association [84] Rees, E. and Owen, M.J. (2020) Translating insights from neuropsychiatric
study in autism. Translat. Psychiatry 10, 1–12. genetics and genomics for precision psychiatry. Genome Med. 12, 1–16.
[67] Luo, Y. et al. (2020) A multidimensional precision medicine approach identifies [85] Karczewski, K. et al. (2021) Systematic single-variant and gene-based association
an autism subtype characterized by dyslipidemia. Nature Med. 26, 1375–1379. testing of 3,700 phenotypes in 281,850 UK Biobank exomes. Cell Genomics 2
[68] Zou, J. et al. (2019) A primer on deep learning in genomics. Nature Genet. 51, 100168.
12–18. [86] Sweeney, N.M. et al. (2021) Rapid whole genome sequencing impacts care and
[69] Kalinin, A.A. et al. (2018) Deep learning in pharmacogenomics: from gene resource utilization in infants with congenital heart disease. NPJ Genomic Med.
regulation to patient stratification. Pharmacogenomics 19, 629–650. 6, 1–10.
[70] Filipp, F.V. (2017) Precision medicine driven by cancer systems biology. Cancer [87] Smith, A.M. King, J.J. West, P.R. Ludwig, M.A. Donley, E.L. and Burrier, R.E.,
Metastasis Rev. 36, 91–108. et al (2019) Amino acid dysregulation metabotypes: potential biomarkers for
[71] Sammut, S.-J. et al. (2021) Multi-omic machine learning predictor of breast diagnosis and individualized treatment for subtypes of autism spectrum
cancer therapy response. Nature 1–10. disorder. Biol. Psychiatry 85, 345–354.
[72] Parsons, J. and Francavilla, C. (2020) Omics approaches to explore the breast [88] Wolfers, T. et al. (2019) From pattern classification to stratification: towards
cancer landscape. Front. Cell Dev. Biol. 7, 395. conceptualizing the heterogeneity of Autism Spectrum Disorder. Neurosci.
[73] Guney, E. and Athie, A. (2021) A needle for Alzheimer’s in a haystack of claims Biobehav. Rev. 104, 240–254.
data. Nature Aging 1, 1083–1085. [89] Yap, I.K. et al. (2010) Urinary metabolic phenotyping differentiates children
[74] Hampel, H. et al. (2017) A precision medicine initiative for Alzheimer’s disease: with autism from their unaffected siblings and age-matched controls. J.
the road ahead to biomarker-guided integrative disease modeling. Climacteric Proteome Res. 9, 2996–3004.
20, 107–118. [90] Smith, A.M. et al. (2020) A metabolomics approach to screening for
[75] Torres, C. and Grippo, P.J. (2018) Pancreatic cancer subtypes: a roadmap for autism risk in the children's autism metabolome project. Autism Res. 13,
precision medicine. Ann. Med. 50, 277–287. 1270–1285.
[76] Wesolowska-Andersen, A. et al. (2022) Four groups of type 2 diabetes contribute [91] Bhat, S. et al. (2014) Autism: cause factors, early diagnosis and therapies. Rev.
to the etiological and clinical heterogeneity in newly diagnosed individuals: An Neurosci. 25, 841–850.
IMI DIRECT study. Cell Rep. Med. 100477. [92] Coe, B.P. et al. (2019) Neurodevelopmental disease genes implicated by de novo
[77] Gottesman, I.I. and Gould, T.D. (2003) The endophenotype concept in mutation and copy number variation morbidity. Nature Genet. 51, 106–116.
psychiatry: etymology and strategic intentions. Am. J. Psychiatry 160, 636–645. [93] Ruzzo, E.K. et al. (2019) Inherited and de novo genetic risk for autism impacts
[78] Guney, E. et al. (2016) Network-based in silico drug efficacy screening. Nature shared networks. Cell 178, 850–866.e26.
Commun. 7, 1–13. [94] Satterstrom, F.K. et al. (2019) Autism spectrum disorder and attention deficit
[79] Aguirre-Plans, J. et al. (2018) Proximal pathway enrichment analysis for hyperactivity disorder have a similar burden of rare protein-truncating variants.
targeting comorbid diseases via network endopharmacology. Pharmaceuticals Nature Neurosci. 22, 1961–1965.
11, 61.

www.drugdiscoverytoday.com 11