Hi, I’m Shraddha Piparia — Computational Biologist with a PhD in computer science 👋

I build machine learning models and analysis pipelines for genomics, proteomics and single-cell data, mostly on questions of disease heterogeneity and treatment response in asthma and long COVID.

Computational Biology + ML | Genomics, Proteomics & Single-Cell Multi-omics

Personal site · Publications · Write-ups

ML/DL: PyTorch · Representation Learning · VAEs · Transformers · SHAP · Clinical NLP

Computational Biology: GWAS · Disease heterogeneity · Single-cell · Multi-omics · UK Biobank · Olink

Engineering: Nextflow · Docker · Spark · SLURM · GitHub Actions

Does a nonlinear encoder trained on genotypes alone capture more phenotype-relevant signal from an LD block than PCA at the same dimensionality, and is any gain specific to asthma regions?

• Per-block β-VAE with an ordinal decoder, compared against per-block PCA with matched latent dimension, across 464 blocks from asthma and control regions.
• Cross-block embedding and phenotype evaluation (held-out partial R², primary phenotype IgE) is under way.

Do children with neurological long COVID have a distinct inflammatory plasma protein profile, and does it hold in adults?

• Olink NPX data from 111 children at two centers. Defined the neurological long COVID group and the comparator design; handled center confounding with ComBat while keeping the primary comparison within a single site.
• Candidate proteins (FGF21, TNFRSF11B, IL17C, CCL2) tested with permutation-based inference; tested replication in more than 50,000 UK Biobank adults using WHO-aligned symptom definitions and a Spark pipeline.
• Manuscript submitted (2026).

Does adding chromatin accessibility improve how well cell identity and state are resolved in PBMCs compared with RNA alone?

• RNA and ATAC processed separately, then combined with weighted nearest neighbor integration in Seurat/Signac. The first ATAC LSI component tracked sequencing depth (ρ ≈ 0.92), so it was excluded from integration.
• Clusters checked against technical QC metrics and annotated with SingleR plus canonical markers.
• Next: motif activity (chromVAR), gene activity, and peak-to-gene links.

Do clinically defined asthma endotypes have genetic associations that a case-control GWAS misses?

• Defined endophenotypes from lung function and treatment response with PCA, then tested variants with ANCOVA-style models. Discovery in CAMP, validation in GACRS. Nextflow pipeline for SLURM.
• Piparia S, et al. A principal component analysis-based endophenotype definition for change in lung function and inhaled corticosteroid treatment response in childhood asthma. Respiratory Research 26:351 (2025). DOI
• Piparia S, et al. A categorical ANCOVA approach to severity endophenotype-specific GWAS in childhood asthma. Journal of Personalized Medicine 16(1) (2026). DOI

Which circulating miRNAs modify how children with asthma respond to inhaled corticosteroids?

• miRNA × treatment interaction models in childhood asthma cohorts; identified miR-584-5p and miR-205-5p as a candidate modifiers of corticosteroid resistance.
• Piparia S, et al. MicroRNA-584-5p as a key modulator of inhaled corticosteroid resistance in asthma. AJRCCM 211 Suppl 1 (2025). DOI

Can the text of chest X-ray impressions alone identify children with COVID-19?

• NLP features from more than 2,500 radiology impressions, random forest classifier, SHAP for interpretation.
• Piparia S, Defante A, Tantisira K, Ryu J. Using machine learning to improve our understanding of COVID-19 infection in children. PLOS ONE 18(2): e0281666 (2023). DOI

A modular template covering read QC, alignment, variant calling, GWAS, polygenic risk scores and eQTL lookup.