X-Admix: An Interpretable Multimodal Cross-Attention Framework for Integrating Genotype, Local Ancestry, and Social Drivers of Health in Admixed African American Populations
- 1University of Arkansas, Fayetteville
Abstract
Disease risk in admixed human populations is shaped by interactions among geno-type, locus-specific ancestry, and the social environment, but predictive frameworks rarely model these three modalities jointly. We introduce X-Admix, an interpretable multimodal framework integrating genotype, local ancestry, and social drivers of health through structured pairwise cross-attention streams, softmax-gated fusion, and a Random Forest classifier. Unlike concatenation-based fusion, these directed streams learn conditional representations in which genotype is contextualized by local ancestry and social drivers of health, and local ancestry by social drivers of health. Leave-one-stream-out ablation decomposes predictive performance and top cross-modal pair candidates. Applied to 240 African American children with severe asthma from the BIG dataset, X-Admix predicted inhaled-corticosteroid response with mean area under the receiver operating characteristic curve 0.771 {+/-} 0.072 across 10-fold cross-validation, whereas ridge logistic baselines performed near chance. This performance pattern replicated in 666 African American adults from the All of Us dataset under matched inclusion criteria. On the BIG dataset, a two-tier consensus pipeline yielded 932 top cross-modal feature pairs whose stream dependencies decomposed into stream-independent (14.7%), single-stream-conditional (28.7%), multi-stream-dependent (33.5%), and all-stream-dependent (23.2%). Without the genotype-local-ancestry stream, performance is unchanged, yet the top cross-modal pairs identified change, indicating that a streams contribution to interpretation and to prediction are separable: a stream redundant for prediction can still define the candidate interactions carried forward for discovery. To our knowledge, X-Admix is the first framework to jointly model the three data modalities via cross-attention in admixed cohorts, yielding an interpretable catalog of candidate interactions underlying inhaled-corticosteroid non-response.
Author SummaryChildren and adults with asthma who share the same diagnosis--and even the same genetic variants -often respond very differently to inhaled steroid medications. In people of mixed ancestry, this variation reflects at least three things acting together: the genetic variants a person carries, the ancestral origin of the surrounding stretch of their genome, and the social and environmental conditions in which they live. Most predictive models treat these factors separately or simply add them together, which can hide how one factor changes the meaning of another. We built a framework, X-Admix, that instead lets each factor provide context for the others, so that a genetic variant can carry different information depending on the ancestry of its surrounding genomic region and a persons environment. In African American children, and again in an independent group of African American adults, we found that X-Admix identified who would not respond to inhaled steroids more accurately than standard models built from the same information. We also obtained a ranked list of gene-ancestry-environment relationships, for example, air-pollution exposure acting together with immune-related genes-that suggest why treatment response varies and that can be tested directly in future studies.
Related
- HAFI-DTA: an interpretable multimodal framework for drug-target affinity prediction based on hierarchical attention guidance and feature iterative interaction
- Application of a multimodal model integrating optical coherence tomography angiography (OCTA) and radiomics for the identification of high-risk Nonarteritic anterior ischaemic optic neuropathy (NAION) populations and for distinguishing its acute phase.
- A federated attention-based stacked LSTM framework for interpretable malaria diagnosis under simulated non-IID federated conditions
- An Interpretable Machine Learning Framework Integrating Multi-Window Environmental Exposure for Hypertension Risk Assessment.
- Interpretable attention-deficit hyperactivity disorder diagnosis from functional magnetic resonance imaging via hybrid convolutional neural network–transformer learning and multi-level graph-based comprehensibility
BibTeX
@misc{tahmin2026xadmix,
title = {{X-Admix: An Interpretable Multimodal Cross-Attention Framework for Integrating Genotype, Local Ancestry, and Social Drivers of Health in Admixed African American Populations}},
author = {N. Tahmin and L. K. Chinthala and T. B. Mersha and R. L. Davis and A. Khojandi},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.03.26362179},
url = {https://www.medrxiv.org/content/10.64898/2026.09.03.26362179v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - X-Admix: An Interpretable Multimodal Cross-Attention Framework for Integrating Genotype, Local Ancestry, and Social Drivers of Health in Admixed African American Populations
AU - N. Tahmin
AU - L. K. Chinthala
AU - T. B. Mersha
AU - R. L. Davis
AU - A. Khojandi
PY - 2026
DA - 2026/09/03
PB - medRxiv
DO - 10.64898/2026.09.03.26362179
UR - https://www.medrxiv.org/content/10.64898/2026.09.03.26362179v1
AB - Disease risk in admixed human populations is shaped by interactions among geno-type, locus-specific ancestry, and the social environment, but predictive frameworks rarely model these three modalities jointly. We introduce X-Admix, an interpretable multimodal framework integrating genotype, local ancestry, and social drivers of health through structured pairwise cross-attention streams, softmax-gated fusion, and a Random Forest classifier. Unlike concatenation-based fusion, these directed streams learn conditional representations in which genotype is contextualized by local ancestry and social drivers of health, and local ancestry by social drivers of health. Leave-one-stream-out ablation decomposes predictive performance and top cross-modal pair candidates. Applied to 240 African American children with severe asthma from the BIG dataset, X-Admix predicted inhaled-corticosteroid response with mean area under the receiver operating characteristic curve 0.771 {+/-} 0.072 across 10-fold cross-validation, whereas ridge logistic baselines performed near chance. This performance pattern replicated in 666 African American adults from the All of Us dataset under matched inclusion criteria. On the BIG dataset, a two-tier consensus pipeline yielded 932 top cross-modal feature pairs whose stream dependencies decomposed into stream-independent (14.7%), single-stream-conditional (28.7%), multi-stream-dependent (33.5%), and all-stream-dependent (23.2%). Without the genotype-local-ancestry stream, performance is unchanged, yet the top cross-modal pairs identified change, indicating that a streams contribution to interpretation and to prediction are separable: a stream redundant for prediction can still define the candidate interactions carried forward for discovery. To our knowledge, X-Admix is the first framework to jointly model the three data modalities via cross-attention in admixed cohorts, yielding an interpretable catalog of candidate interactions underlying inhaled-corticosteroid non-response. Author SummaryChildren and adults with asthma who share the same diagnosis--and even the same genetic variants -often respond very differently to inhaled steroid medications. In people of mixed ancestry, this variation reflects at least three things acting together: the genetic variants a person carries, the ancestral origin of the surrounding stretch of their genome, and the social and environmental conditions in which they live. Most predictive models treat these factors separately or simply add them together, which can hide how one factor changes the meaning of another. We built a framework, X-Admix, that instead lets each factor provide context for the others, so that a genetic variant can carry different information depending on the ancestry of its surrounding genomic region and a persons environment. In African American children, and again in an independent group of African American adults, we found that X-Admix identified who would not respond to inhaled steroids more accurately than standard models built from the same information. We also obtained a ranked list of gene-ancestry-environment relationships, for example, air-pollution exposure acting together with immune-related genes-that suggest why treatment response varies and that can be tested directly in future studies.
ER -