Spatial representation learning
Learn task-relevant maps of unordered tabular and molecular measurements so local structure becomes available to vision models and human inspection.
Postdoctoral Scholar · Stanford School of Medicine
Interpretable AI for high-dimensional biology.
I develop learning systems that expose structure in biomedical tables, molecular networks, and single-cell data—so predictive models can also support biological reasoning.
My research connects representation learning, computational genomics, and computer vision, with applications in liquid biopsy, multimodal cancer analysis, graph learning, and single-cell biology.
DynomapTables learn their own spatial organization.
Features move · neighborhoods emerge · attributions return to biology
6
Open research platforms
Cartography · graph AI · single-cell models
4
Stanford OTL disclosures
Spatial representations · graph AI
20+
Scholarly works
Biomedical AI · genomics · vision
3
Recent 2026 venues
AACR · AAPM · MICCAI AMAI
Research program
Learning representations that scientists can interrogate.
My long-term goal is to build biomedical AI that does not treat representation as a fixed preprocessing choice. The model should learn or preserve biologically meaningful structure—and return its evidence in a form that researchers can examine.
Research thesis
Predictive performance and scientific interpretability can arise from the same representation.
Read the detailed research programGraph and foundation models
Use topology as a transferable language across heterogeneous networks, patient-specific molecular graphs, and large-scale biological systems.
Translational cancer AI
Develop data-efficient methods for liquid biopsy and multimodal cancer analysis that remain testable under realistic clinical constraints.
Evidence returned to biology
Connect predictions back to genes, pathways, spatial neighborhoods, and structural counterfactuals rather than stopping at a model score.
Open research
From difficult data to inspectable maps.
Six connected projects explore the same central idea at different biological scales: representation is not a formatting detail—it can be the mechanism that makes learning and interpretation possible.
Lead work · Biomedical tabular AI
Dynomap
Learns where features belong while learning the prediction task, converting unordered clinical and molecular tables into continuous spatial maps.
Up to 18%higher multiclass cancer-subtype accuracy in liquid-biopsy experiments
Lead work · Biological networks
Graph2Image
Transforms very large biological graphs into image sets, giving vision models global context while making node-level evidence directly visible.
>1 billionnodes can be analyzed on a personal computer in reported experiments
Lead work · Transferable graph AI
Graph Foundation Model
Learns a shared language of node structural roles on non-biological graphs, then transfers that topology knowledge into unseen biological networks.
95.5%SagePPI mean ROC–AUC, 21.8 points above the strongest supervised baseline
Collaborative work · Knowledge-guided graph learning
Graph-in-Graph
Builds one pathway-structured molecular graph per patient, combining curated gene relationships with patient-specific expression for data-efficient clinical prediction.
+49 pointsmacro-F1 improvement on the most challenging prostate-cancer task
Collaborative work · Single-cell biology
scVision
Places co-expressed genes near one another, renders each cell as an image, and learns a reusable representation through masked-image pretraining.
72 millionhuman cells used for self-supervised pretraining
In development · Structural language
BioGlyph
Compiles a graph into named structural roles, measured evidence, and counterfactual descriptions so a frozen language model can reason about topology.
In developmentmanuscript in preparation · deterministic, interpretable graph-to-language reasoning
Visit project
Manuscript in preparation
Selected manuscripts & preprints
Current work.
Dynomap · 2026
Vision-Based Deep Learning of Biomedical Tabular Data via Self-Supervised Cartographic Representation
Nature Communications · Under review
Graph Foundation Model · 2026
Language-Encoded Structural Topology Enables Generalizable Foundation Models for Graph-Structured Data
Nature Computational Science · Under review
Graph2Image · 2026
Transformation of Biological Networks into Images via Semantic Cartography for Visual Interpretation and Scalable Deep Analysis
Nature Biomedical Engineering · Under review
Graph-in-Graph · 2026
Knowledge Graph Modulated Deep Learning for Limited-Sample Clinical Data Analysis
Nature Biomedical Engineering · Preparing for submission
scVision · 2026
A Vision Foundation Model for Single-Cell Biology via Spatial Gene Cartography
Nature · Preparing for submission
News & recognition
Recent milestones.
Workshop
↗
Poster abstract accepted to the 5th Workshop on Applications of Medical Artificial Intelligence at MICCAI 2026.
The workshop will be held in conjunction with MICCAI in Strasbourg.
Fellowship
Awarded the Stanford School of Medicine Dean’s Fellowship.
Supporting postdoctoral research in interpretable biomedical AI.
Presentation
Presented multimodal molecular and medical-imaging research at the AAPM Annual Meeting.
Oral presentation in Vancouver, Canada.
Funding
Received a Google Cloud research grant for scalable biomedical AI.
Cloud support for large-scale model development and evaluation.
AACR
↗
Presented image-based multimodal AI for integrating imaging and genomics in cancer research.
AACR Annual Meeting abstract 5490.
Team award
↗
Stanford Cancer Institute Innovation Award supports our deep-learning framework for liquid biopsy.
Contributing researcher on the Islam Lab project with Stanford collaborators.
In the media
Making trustworthy AI part of the public conversation.
Interviews and profiles on explainable AI, cancer research, and the path from deep-learning methods to high-stakes biomedical use.
College of Arts and Science, University of Saskatchewan · October 2025
Unpacking the black box of AI
A profile of my work on explainable AI and interpretable cancer-detection models at Stanford.
Read the profile ↗ CBC Canada · Video interviewMultimodal AI for cancer research and medicine
A video conversation about combining imaging, genomic, and clinical information for more robust and interpretable cancer analysis and precision medicine.
Watch the CBC interview ↗ The Morning Edition — Saskatchewan, CBC Radio · October 2025This U of S grad wants to ensure you can trust AI
A radio interview on why medical AI must be understandable—not merely accurate—before it can be trusted in cancer diagnosis.
Listen on CBC Radio ↗ University of Saskatchewan News · February 2022Diving into deep learning
A research feature on opening the black box of deep learning for plant phenotyping, food security, and other high-stakes applications.
Read the feature ↗Teaching, mentoring & service
Building people and research communities.
My academic work includes classroom teaching, hands-on research mentoring, and more than twenty peer reviews across biomedical AI, computer vision, and computational biology.
From lecturer to course coordination
University teaching across AI, computer vision, algorithms, and computing—including coordinating 13 teaching assistants for a 250-student course.
Research training across career stages
Mentoring PhD, master’s, undergraduate, and clinical research trainees in biomedical AI, single-cell biology, and graph learning.
Review service across biomedical AI and computing
Reviewer for MICCAI, ACM Transactions on Computing for Healthcare, IEEE TETCI and IEEE Access, Neurocomputing, Frontiers journals, BMC Plant Methods, IEEE BIBM, ICCABS, and CCCG.
Collaboration network
Research built across institutions, disciplines, and sectors.
Explore selected current and longstanding collaborations in biomedical AI, computational biology, cancer research, genomics, plant science, and translational technology.
Select a region or use the arrow keys
Select to open a region
Select to view collaborators
3 regions · academic, government, and industry
Contact
Let’s build AI that scientists can inspect—and trust.
I welcome conversations about faculty opportunities, research collaborations, student mentorship, and translational biomedical AI.
sakib@stanford.edu