Jo Lab

The Jo Lab develops open-source deep learning tools and methods for large-scale genomic and multi-omics analysis in Alzheimer’s disease.

Taeho Jo, PhD, is an assistant professor in the Department of Radiology and Imaging Sciences at Indiana University School of Medicine. His research develops deep learning methods for the early detection of Alzheimer’s disease, integrating whole-genome sequencing (WGS), neuroimaging (tau PET, MRI), proteomics and metabolomics. He conducts AI research at the Indiana Alzheimer’s Disease Research Center and the Center for Neuroimaging, and participates in the AI for Alzheimer’s Disease (AI4AD) consortium.

Established in 2024, the Jo Lab builds open-source deep learning tools for large-scale genomic and multi-omics analysis in Alzheimer’s disease. The lab’s computational frameworks, spanning CNN, transformer, and ensemble architectures, are designed to identify genetic risk variants, discover blood-based biomarkers, and support precision medicine. All tools are publicly available on GitHub and as interactive web platforms at jolab.ai.

Explore the Jo Lab website

AI-Driven Whole-Genome Sequencing Analysis for Alzheimer’s Disease

The lab develops deep learning methods to identify AD-associated genetic variants from whole-genome sequencing data containing over 12 million SNPs per individual. SWAT (Sliding Window Association Test) uses a CNN-based sliding window to scan the genome while preserving positional information of variants, achieving an AUC of 0.82 (Briefings in Bioinformatics, 2022). Deep-Block extends this by incorporating linkage disequilibrium (LD) to partition the genome into biologically meaningful blocks, then applies transformer-based attention to detect risk loci. Applied to the ADSP dataset (n=7,416), it improved classification accuracy by 17-22% (AD TRCI, 2025).

DuAL-Net combines local genomic context analysis with global functional annotations centered on the APOE region, using a TabNet + Random Forest ensemble (CSBJ, 2026). TrUE-Net addresses the gap of confidence estimation in genomic predictions by integrating a transformer with Monte Carlo Dropout and a Random Forest classifier, enabling distinction between high-confidence and uncertain predictions (Briefings in Bioinformatics, 2025).

Workflow diagram for whole-genome sequencing analysis using machine learning. Genomic data undergoes quality control, linkage disequilibrium block detection, and automated imputation. Multiple TabNet encoder models with feature transformers perform attentive feature selection and key block decisions. The final stage combines TabNet outputs with a random forest model for identifying putative impact SNPs.

[Figure 1] Briefings in Bioinformatics 2022 (Deep-Block/SWAT pipeline overview)
Multi-stage deep learning framework for WGS-based Alzheimer’s disease risk variant identification

Diagram illustrating a multi-layer framework for SNP prioritization across the whole genome sequence. Horizontal colored tracks represent annotation categories including variant annotation features, clinical significance, variant effect, and gene biotype. A sliding window–based approach partitions the genome into sequential windows. Each layer highlights a decreasing number of selected SNPs, visually summarized at the right side. The bottom axis shows the full genome sequence divided into windows.

[Figure 2] CSBJ 2026 DuAL-Net architecture
DuAL-Net: local SNP window + global functional annotation

Deep Learning for Neuroimaging-Based Alzheimer’s Detection

The lab applies 3D CNN models to tau PET scans for AD classification, achieving 90.8% accuracy in distinguishing AD from cognitively normal controls. Integrated with Layer-wise Relevance Propagation (LRP), the framework generates interpretable maps of the brain regions most informative for classification, consistent with known tau pathology patterns in the medial temporal lobe and posterior cortex (BMC Bioinformatics, 2020).

Schematic of a 3D convolutional neural network applied to Tau-PET brain images. The diagram shows stacked brain volumes entering multiple 3D convolution, batch normalization, max-pooling, and dropout layers with increasing feature maps. A fully connected layer generates predictions. A secondary pathway illustrates model validation, accuracy testing, and generation of relevance heatmaps using layer-wise relevance propagation.

[Figure 3] BMC Bioinformatics 2020, 3D CNN + LRP tau PET visualization
3D CNN applied to tau PET with Layer-wise Relevance Propagation highlighting informative brain regions for AD classification

Blood-Based Biomarker Discovery Through Proteomics and Metabolomics

In metabolomics, c-SWAT (Circular Sliding Window Association Test) adapts the SWAT architecture for serum-based lipidome data. Applied to 781 lipid species from ADNI, c-SWAT achieved up to 80.8% accuracy in classifying cognitively normal, MCI, and AD groups (eBioMedicine, 2023).

A study published in Alzheimer’s & Dementia (2025) analyzed longitudinal plasma proteomics data and identified six proteins that predict incident AD with an AUC of 0.76, providing a minimally invasive approach to early detection before cognitive symptom onset.

Two-panel visualization of metabolomics analysis. Panel (a) shows a circular SWAT diagram with concentric rings indicating sliding windows, WGCNA modules, and metabolite groupings. Panel (b) presents a circular chord diagram linking metabolite categories to multiple groups, with multicolored connecting lines representing associations between metabolites and network modules.

[Figure 4] eBioMedicine 2023 (c-SWAT architecture or classification results)
Circular Sliding Window Association Test (c-SWAT) applied to serum lipidomics for AD classification

Precision Medicine Through Multimodal AI Integration

The lab’s overarching goal is to integrate genomics, neuroimaging and blood-based biomarkers through AI for individualized predictions of AD progression. This multimodal approach captures complementary layers of disease biology: inherited genetic risk from WGS, in vivo neuropathology from tau PET, and peripheral biochemical signatures from proteomics and metabolomics.

All tools are released as open-source software on GitHub, with web-based platforms available at jolab.ai.

Flowchart showing an integrated multi-omics analysis pipeline. Genomic SNP data, neuroimaging inputs, and metabolomics data are processed through deep learning models including convolutional neural networks, attention mechanisms, SWAT-CNN, transformers, and deep-block architectures. Outputs emphasize long-range dependencies and produce connected circular and network visualizations representing associated biological features in Alzheimer’s disease.

[Figure 5] genomics + neuroimaging + proteomics/metabolomics converging through AI
Integrative AI framework combining genomic, neuroimaging, and multi-omics data for precision medicine in AD

Recent Publications

Eun Hye Lee, Taeho Jo*, “DuAL-Net: A Dual-Network Approach for Alzheimer’s Disease Risk Prediction Using APOE-Centered Regional WGS Data,” Computational and Structural Biotechnology Journal, 2026.

Taeho Jo*, Eun Hye Lee, “Uncertainty-Aware Genomic Classification of Alzheimer’s Disease: A Transformer-Based Ensemble Approach with Monte Carlo Dropout,” Briefings in Bioinformatics, 2025.

Eun Hye Lee, ..., Andrew J. Saykin*, Taeho Jo*, and Kwangsik Nho*, “Longitudinal plasma proteomics: relation to incident Alzheimer disease and biomarkers,” Alzheimer’s & Dementia, 2025.

Taeho Jo*, Paula Bice, Kwangsik Nho, Andrew J. Saykin, “Linkage Disequilibrium-Informed Deep Learning Framework to Identify Genetic Loci for Alzheimer’s Disease Using Whole Genome Sequencing Data,” Alzheimer’s & Dementia: TRCI, 2025.

Taeho Jo, Junpyo Kim, ..., Andrew J. Saykin, and Kwangsik Nho, “Novel circular-SWAT for deep learning based diagnostic classification of Alzheimer’s disease: Application to metabolome data,” eBioMedicine, 2023.

Taeho Jo, Kwangsik Nho, Paula Bice, and Andrew J. Saykin, “Deep learning-based identification of genetic variants: Application to Alzheimer’s disease classification,” Briefings in Bioinformatics, 2022.

Google Scholar

Current Research Funding

Jo Lab

AARG-22974053 (Alzheimer’s Association), PI: Jo
A Dual-Deep Learning AI Strategy to Identify Tau-associated Genetic Variants in Alzheimer’s Disease

Collaborative

U01 AG068057 (NIH/NIA), PIs: Thompson & Saykin, Co-I: Jo
Ultra-scale Machine Learning to Empower Discovery in Alzheimer’s Disease Biobanks

U01 AG072177 (NIH/NIA), PI: Saykin, Co-I: Jo
KBASE2: Korean Brain Aging Study, Longitudinal Endophenotypes and Systems Biology

P30 AG072976 (NIH/NIA), PI: Saykin, Co-I: Jo
Indiana Alzheimer’s Disease Research Center (IADRC)

U19 AG024904 (NIH/NIA), PI: Weiner, Co-I: Jo
The Alzheimer’s Disease Neuroimaging Initiative (ADNI-4)

Members

Principal Investigator

Taeho Jo, PhD

Assistant Professor of Radiology & Imaging Sciences

Read Bio

Visiting Scholar

Donghyun Lee, MD

Previous affiliation: Clinical Assistant Professor, Hallym University College of Medicine; Asan Medical Center

Research Intern

Leonard Lee, MS

MS in AI, Johns Hopkins University
BS in Computer Science, Purdue University

Research Volunteer

Aryan Rai

Medical Student, IU School of Medicine