dp.Devansh Pancholi

Computational biology · Multi-omics · AI

DNA. RNA. Cells.
Connected by computation.

I’m Devansh, a computational biologist applying computational methods to understand cancer across molecular scales—how genomic alterations connect with RNA programs, epigenetic regulation, cellular states, and disease behavior.

My research spans single-cell and bulk transcriptomics, somatic-variant analysis, cfDNA biomarkers, and developing spatial integration. I’m particularly interested in how molecular and cellular programs change after treatment, and how these shifts relate to resistance, disease progression, and survival. Alongside this work, I develop reproducible multi-omics workflows and adapt AI models for RNA structure prediction, connecting biological questions with interpretable analysis.

Explore my research Download résumé
M.S. Bioinformatics · Saint Louis UniversityOpen to computational research roles
Single-cell & spatial transcriptomicsDNA & epigenetic regulationStructural biology & AIMulti-omics & biomarkers

01 / Selected research

Biological questions.
Computational evidence.

From tumor ecosystems to molecular structure: projects grounded in interpretable analysis, cohort context, and reproducible workflows.

02 / Single-cell genomics72

A multi-sample
breast-cancer atlas

Integrated 72 scRNA-seq samples spanning normal tissue, ER-positive, HER2-positive, triple-negative, and metastatic disease.

  • Performed multi-sample QC, batch-aware integration, clustering, and cell-type annotation.
  • Examined malignant, immune, and stromal populations through differential-expression and pathway analysis.
scRNA-seqScanpyCell-state heterogeneity
Explore repository
Project figures
Breast-cancer single-cell atlas
Breast-cancer single-cell atlas

Project overview: 72 tissue samples spanning normal, ER-positive, HER2-positive, triple-negative, and metastatic contexts. The overview graphic summarizes the project; it is not a measured UMAP.

Open full-size image ↗
Cell-type marker expression — panel 1
Cell-type marker expression — panel 1

Marker-expression review across the displayed epithelial, macrophage, fibroblast, and endothelial groups. Dot size represents the fraction of expressing cells; color represents mean expression.

Open full-size image ↗
Cell-type marker expression — panel 2
Cell-type marker expression — panel 2

Marker-expression review including T-cell and myeloid programs. Compare canonical markers across groups to assess annotation consistency.

Open full-size image ↗
Cell-type marker expression — panel 3
Cell-type marker expression — panel 3

Additional marker-expression panel for reviewing assigned cell labels. Mixed lineage-marker signals warrant follow-up before biological interpretation.

Open full-size image ↗
03 / Liquid-biopsy epigenomics

Reading cancer signals
in plasma cfDNA

Combined colorectal-cancer 5hmC profiling and low-pass WGS fragmentomics across discovery and independent validation cohorts at WashU.

  • Quantified gene-body 5hmC, fragment lengths, end motifs, and nucleosome/TF-binding-site-associated features.
  • Evaluated progression-associated signals and pre-analytical differences between EDTA and Streck collection.
5hmCFragmentomicsRegulatory analysis
Explore repository
Project outputs
Hallmark enrichment: PD vs NED
Hallmark enrichment: PD vs NED

Normalized enrichment scores for the PD versus NED comparison, with FDR < 0.05 as labeled in the supplied figure.

Open full-size image ↗
Hallmark enrichment: CRC vs non-cancer
Hallmark enrichment: CRC vs non-cancer

Pathway-level comparison grouped by biological process. The displayed significance threshold is FDR < 0.05.

Open full-size image ↗
Coverage around CDX2 sites
Coverage around CDX2 sites

Coverage profiles for CRC (n = 66) and non-cancer (n = 147), as labeled in this figure. These counts apply to this panel’s analysis.

Open full-size image ↗
CDX2 genomic tracks
CDX2 genomic tracks

Supplied locus-level comparison of CRC and non-cancer tracks around CDX2. Track heights should be interpreted with their original scaling.

Open full-size image ↗
Fragment-length signal
Fragment-length signal

Original project figure comparing short-to-long fragment ratios. An exploratory molecular feature, not a clinical diagnostic.

Open full-size image ↗
04 / Somatic cancer genomics

From tumor–normal reads
to variant interpretation

A reproducible breast-cancer WGS workflow for somatic SNV and indel detection, filtering, annotation, and pathway-level interpretation.

  • Connected alignment and BAM processing with GATK Mutect2 somatic variant calling.
  • Used SnpEff annotations to prioritize candidate cancer-associated genes for biological interpretation.
BWA-MEMMutect2SnpEff
Explore repository
Project figures
Tumor–normal somatic variant analysis
Tumor–normal somatic variant analysis

Project overview: alignment, somatic calling, functional annotation, and cancer-gene prioritization. The overview artwork summarizes the workflow.

Open full-size image ↗
Prioritized variant candidates — sample 11871
Prioritized variant candidates — sample 11871

Candidate variants ranked by tumor log-odds (TLOD), including PIK3CA p.His1047Arg and CDH1 p.Arg335*. “11871” identifies the sample. TLOD measures evidence for a variant, not whether it is a functional driver.

Open full-size image ↗
Cancer-gene variant consequences
Cancer-gene variant consequences

Gene-by-effect matrix from the project repository, summarizing annotated consequences across selected cancer-associated genes. View repository figure.

Open full-size image ↗
05 / RNA structural informatics

RNAMotifDB:
sequence to template

An RNA motif database and template-search workflow connecting sequence, secondary structure, and tertiary-template information.

  • Built motif extraction, sequence grouping, RMSD-based clustering, and representative-template selection workflows.
  • Connected template search with motif-guided RNA three-dimensional structure assembly.
Python / C++Structural alignmentRNA motifs
Explore repository
View the template-search workflow
RNAMotifDB template-search example showing motif decomposition and structural template matches
Original RNAMotifDB project illustration: motif decomposition and template search.
06 / Machine learning & RNA biology

Better structural inputs.
More informative predictions.

Modified OpenFold3 to accept RNA structural templates and integrated synthetic multiple-sequence alignments derived from RNA secondary structure, then benchmarked predictions against baseline models.

  • Developed parallel Nextflow/HPC workflows for preprocessing, inference, PDB generation, and structural evaluation.
  • Achieved over 70% lower C1′ RMSD in selected benchmark cases: for 1P5P, the combined approach reduced RMSD from 22.452 Å to 3.565 Å (approximately 84%). Performance varied across the 22-RNA comparison, with non-improving cases retained.
Modified OpenFold3RNA templates / synthetic MSANextflow
Explore repository

View per-RNA benchmark results ↗

Project outputs
Reference, templates, and sequence-only predictions
Reference, templates, and sequence-only predictions

Both prediction panels are annotated 1.42 Å in this example. This panel illustrates a comparison, not a template-associated improvement.

Open full-size image ↗
RNA structural overlay
RNA structural overlay

A supplied structural rendering for qualitative comparison of RNA conformations. No additional accuracy score is inferred from the image.

Open full-size image ↗
1E7K: baseline and synthetic-MSA comparison
1E7K: baseline and synthetic-MSA comparison

Individual example retained from the original portfolio. This figure is separate from the aggregate benchmark summary in the project description.

Open full-size image ↗

How I work

From a biological question
to a reproducible result.

The same analytical discipline connects my sequencing, biomarker, and structural-modeling projects.

  1. 01

    Define the question

    Identify the biological comparison, appropriate datasets, sample structure, and relevant metadata.

  2. 02

    Prepare & quality-check

    Inspect technical variation; preprocess reads, count matrices, genomic features, or molecular structures.

  3. 03

    Build the representation

    Generate gene programs, regulatory features, variants, structural templates, or learned embeddings.

  4. 04

    Compare & evaluate

    Use statistical analysis, cohort comparisons, interpretable models, or structural benchmarks to test the question.

  5. 05

    Communicate & reproduce

    Document decisions, visualize evidence, retain limitations, and organize code for repeatable analysis.

02 / Research experience

From methods
to meaningful analysis.

Research across cancer transcriptomics, clinical-cohort epigenomics, and machine-learning-enabled RNA biology.

Apr 2026 — PresentSaint Louis University

Bioinformatics Analyst

HNSCC tumor–immune states & multimodal RNA biomarkers

  • Analyzing human HNSCC single-cell data with an emphasis on macrophage-associated immune-response programs.
  • Developing treatment-response comparisons and single-cell-derived signatures for bulk-cohort analysis.
  • Extending the research toward spatial transcriptomic integration and tissue-level immune organization.
Jan 2026 — May 2026Washington University in St. Louis

Research Assistant

Colorectal-cancer epigenomic biomarkers · Advisor: Dr. Maher

  • Integrated plasma cfDNA 5hmC and LP-WGS features across discovery and validation cohorts.
  • Implemented and troubleshot Python/R and HPC workflows for alignment, feature generation, regulatory analysis, and cohort comparisons.
  • Examined baseline molecular profiles in relation to later clinical outcomes while assessing collection-related confounders.
Mar 2025 — May 2026Saint Louis University

Research Assistant

AI-enabled RNA structure prediction · Advisor: Dr. Jie Hou

  • Built RNAMotifDB and Python/C++ workflows for motif extraction, alignment, clustering, and template-guided assembly.
  • Modified nine OpenFold3 source files to support RNA templates and integrated secondary-structure-derived synthetic MSAs for benchmarked structure prediction.
  • Engineered parallel inference and evaluation workflows with structural repair and residue-coordinate mapping.

03 / Technical expertise

A toolkit built
around the question.

Established analysis and engineering experience, with developing methods identified explicitly.

01

Transcriptomics & cell states

Single-cell QC and integration, annotation, differential expression, gene-module scoring, pathway analysis, and bulk-transcriptomic analysis.

ScanpySTARfeatureCountsDESeq2 / edgeR

Developing: spatial integration, cellular neighborhoods, and expression-based cell–cell communication.

02

Genomics & epigenomics

Tumor–normal variant workflows, cfDNA 5hmC, fragmentomics, cohort comparisons, and regulatory-feature interpretation.

BWASAMtoolsGATKSnpEffMACS2Griffin
03

Machine learning & RNA

Biological feature engineering, explainable models, RNA sequence/structure representations, and prediction benchmarking.

scikit-learnXGBoost / SHAPOpenFold3 modificationDiffusion modelsRiNALMo / ESM-2
04

Reproducible computation

Workflow development, HPC execution, debugging, automation, version control, and clear analytical reporting.

Python / R / C++Bash / SQLLinux / SLURM / LSFNextflow / SnakemakeDocker / Git / AWS

04 / Background

An engineering foundation.
A biological focus.

I trained in biomedical engineering before completing my M.S. in Bioinformatics at Saint Louis University. I’m interested in how molecular measurements reveal cancer biology—and in building the computational workflows that make those measurements interpretable.

My research connects DNA variation, RNA expression, epigenetic regulation, and molecular structure. Cancer and biomarker studies are important applications of this broader interest in multi-omics analysis and biological machine learning.

Read my full résumé
MAY 2026

M.S. Bioinformatics

Saint Louis University, USA

GPA: 3.97

JUNE 2023

B.E. Biomedical Engineering

L.D. College of Engineering, India

GPA: 3.63

Earlier work, still part of the story.

PUBLICATION / IEEE ASIANCON 2023

Smart Stethoscope

IoT-enabled medical-device work connecting biomedical signal acquisition, hardware, and computation.

View publication
APPLIED MACHINE LEARNING

Protein representations & interpretable models

Earlier ESM-2-based classification work using biological embeddings and SHAP interpretation, retained alongside my current cancer research.

Browse GitHub

Research interests

Across molecular layers.
Across biological questions.

Areas I want to keep building in, combining established experience with new analytical directions.

05 / What’s next

Let’s turn complex data
into biological insight.

I’m interested in computational biology and bioinformatics research spanning single-cell and spatial transcriptomics, genomics, epigenetic regulation, biomarkers, structural biology, and machine learning.

pancholidevansh29@gmail.com
GitHub ↗LinkedIn ↗Download résumé ↓