A Deep Learning Algorithm Platform to Predict Autism Diagnosis and Subtypes

A Deep Learning Algorithm Platform to Predict Autism Diagnosis and Subtypes by Integrating Clinical, Cognitive, Imaging, Gut Microbiome, and Metabolome Data

This is the first human study on ASD microbiome with robust methodologies: prospective and sibling designs, metagenomics profiles, establishing an ASD multi-dimensional databank (clinic, behavior, neurocognition, brain imaging, metabolomics, and microbiome) collected using the same methodology and genetic biology simultaneously, and developing a deep learning platform for ASD diagnosis and prevention. With the accomplishment of this project, we anticipate establishing a web application for clinical and academic use. Our findings will further advance the knowledge in the pathogenetic mechanisms of ASD to enhance early detection, diagnosis, and treatment, subsequently contributing to precision medicine.

Due to the high prevalence (1% in Taiwan), long-lasting impairment, unclear etiologies, and a lack of effective detection, prevention, and biological treatment, autism spectrum disorder (ASD) has been prioritized for biomarker, mechanism, and treatment research. Recently the gut-brain-axis has been proved, mainly with animal models, to be altered in psychiatric disorders and notably in ASD. With PI Gau's long-term achievement in ASD multi-dimensional research and our preliminary finding of altered gut microbiota in ASD and their unaffected siblings, we propose this 4-year prospective large-scale study with sibling design and multi-dimensional measures (environmental, clinical, cognitive, imaging, gut microbiome, metabolome) to establish a deep learning algorithm platform for predicting ASD and searching potential biomarkers and probiotic treatment for ASD.

Specific Aims:

1. To demonstrate the metagenomics profiles analysis based on the gut microbiome and metabolome of ASD patients, unaffected siblings, and typically developing controls (TDC).
2. To investigate environmental factors such as pregnancy and birth history from the mother's medical records and interviews or national health insurance data, for the microbiome, metagenomics, and brain anatomy and function.
3. To develop a deep learning algorithm platform using the environmental, behavioral/clinical phenotypes, neurocognitive/imaging endophenotypes, and metagenomics profiles to identify microbiota (metagenomics, too) makers and other predictors for ASD diagnosis, subtypes, and level of impairments.
4. To establish a web application based on our deep learning algorithm platform for clinical use to assist medical doctors in diagnosing ASD.

Inclusion Criteria:

• ASD participants are (1) they have a clinical diagnosis of ASD defined by the DSM-5 criteria,1 made by board-certificated child psychiatrists and confirmed by the ADI-R/ADOS; (2) their ages range from 4 to 25; (3) both parents are Han Chinese; (4) they and their parents cooperate with all the assessments and stool and blood collection.

Inclusion Criteria for US and TDC are (1) they do not reach the clinical diagnosis of ASD according to DSM-5 diagnostic criteria and the same criteria as described in the (2), (3), (4) and of Inclusion Criteria for ASD participants.

Exclusion Criteria:

• (1) comorbidity with DSM-5 diagnoses of schizophrenia, schizoaffective disorder, delusional disorder, other psychotic disorders, organic psychosis, schizotypal personality disorder, bipolar disorder, depression, severe anxiety disorders or substance use; (2) comorbidity with neurological or systemic disorders; and (3) having a first degree relative who may have ASD based on family history method assessment (the TDC group).