Projects
Gamified Data Collection & Digital Intervention
The Wall Lab at Stanford is committed to developing AI-enhanced and data backed games that can improve early detection and intervention options for complex conditions that originate in childhood and perpetuate through the life course, including autism and related developmental delays.
AI Model Development
We're on a mission to pioneer early detection and enhance outcomes for children with ASD and related developmental delays. Rooted in our mission we explore all frontiers spanning from detection to improving algorithmic performance and building robust datasets. By embracing these diverse fronts, we're committed to uncovering insights that can make a tangible difference.
Overview on current research:
-
Ensemble Learning
Ensemble learning offers a versatile framework for leveraging the strengths of multiple models. Using ensemble learning approaches we can identify subgroups within the spectrum by clustering individuals based on their characteristics or response patterns, thereby enabling more personalized interventions or treatments.
-
Crowdsourcing for NT data
Limited data set size can compromise the generalizability and robustness of the AI models, particularly in health-critical fields like digital phenotyping that require minimal bias and optimal performance. Crowdsourcing enables one to tap into the collective wisdom and capabilities of a diverse group of individuals, fostering collaboration, and problem-solving on a global scale. It offers a flexible and scalable approach to addressing various challenges.
-
Temporal Adaptation
We introduce Temporal consistency for Test-time adaptation (TempT), a novel method for test-time adaptation on videos through the use of temporal coherence of predictions across sequential frames as a self-supervision signal. TempT is an approach with broad potential applications in computer vision tasks, including facial expression recognition (FER) in videos.
Precision Health
Our lab uses genetic data from families to identify the genes that are involved in autism spectrum disorder. To do this, we develop new computational methods that leverage familial relationships to identify autism risk regions.
-
The genome-wide sibling-pair linkage test
We developed a hidden Markov model for resolving crossovers and shared genetic material (IBD) in WGS data from families. We then developed a genome-wide sibling pair linkage test which uses IBD across large numbers of families to identify autism risk regions. By working with sibling IBD, our method is able to detect genomic regions harboring risk variants, even if the risk variants themselves are not sequenced and do not exhibit linkage disequilibrium with their neighbors. Furthermore, our method can identify parent-of-origin effects which are known to play a role in neurodevelopmental processes. Using crowd-sourced microarray data from 132 autism families, our method identified two autism risk regions which we were able to validate with an independent dataset. We then used variant transmission patterns to demonstrate that alternative splicing in one of the regions likely plays a role in autism risk. This work provides a proof-of-concept, demonstrating that extending family-based linkage analysis into the era of next-generation sequencing has the potential to increase our understanding of genetic risk factors for complex disorders.
Publications:
Paskov K, Chrisman B, Stockham N, Washington PY, Dunlap K, Jung JY, Wall DP. Identifying crossovers and shared genetic material in whole genome sequencing data from families. Genome Research. 2023 Oct 1;33(10):1747-56. -
iHART
In collaboration with the Geschwind Lab and the Hartwell Foundation, we sequenced and analyzed 1,006 families with two or more children with autism. This dataset is an important resource for studying inherited genetic risk factors for autism.
Publications:
Ruzzo EK, Pérez-Cano L, Jung JY, Wang LK, Kashef-Haghighi D, Hartl C, Singh C, Xu J, Hoekstra JN, Leventhal O, Leppä VM. Inherited and de novo genetic risk for autism impacts shared networks. Cell. 2019 Aug 8;178(4):850-66. -
Shapley values
We have repurposed the Shapley value, a game theoretic approach to quantify the marginal contribution of a “player” (in this case, a single nucleotide polymorphism) to the “game outcome” (in this case, prediction of autism).
Publications:
Sun MW, Moretti S, Paskov KM, Stockham NT, Varma M, Chrisman BS, Washington PY, Jung JY, Wall DP. Game theoretic centrality: a novel approach to prioritize disease candidate genes by combining biological networks with the Shapley value. BMC bioinformatics. 2020 Dec;21:1-0.
Sun MW, Gupta A, Varma M, Paskov KM, Jung JY, Stockham NT, Wall DP. Coalitional game theory facilitates identification of non-coding variants associated with autism. Biomedical Informatics Insights. 2019 Mar;11:1178222619832859.
Gupta A, Sun MW, Paskov KM, Stockham NT, Jung JY, Wall DP. Coalitional game theory as a promising approach to identify candidate autism genes. InPACIFIC SYMPOSIUM ON BIOCOMPUTING 2018: Proceedings of the Pacific Symposium 2018 (pp. 436-447).