research
publications
-
towards causal interpretability in deep learning for parkinson's detection from voice data
princeton journal of interdisciplinary research (pjir) · vol. 1, issue 2
aniruth ananthanarayanan*, sudeep senivarapu*, anishsairam murari* (*equal contribution)
benchmarked cnn, mc-dropout, few-shot, and ensemble models for non-invasive parkinson's detection from voice (97.4% accuracy, auc 0.9897 on a held-out test set), then used causal forests to estimate biomarker effect sizes beyond correlation. covered by new scientist, bbc science focus, and sciencealert.
-
computational protocol for organism-level pathogenicity classification from bacterial genomes using machine learning
book chapter · springer nature
sudeep senivarapu, tallon coxe, anishsairam murari, rajeev k. azad
a reproducible end-to-end protocol for classifying bacterial pathogenicity from genome sequences: bacspad and bv-brc dataset construction, then k-mer svms, random forests, and a reverse-complement-aware cnn.
datasets
-
an extended dataset of extracted acoustic features for voice-based detection of parkinson's disease
extended acoustic feature dataset for parkinson's disease voice analysis research.
presentations
-
oral & poster presentation
bmes annual meeting 2025 · selected in the general abstract category, reviewed alongside graduate and postdoctoral submissions
-
poster presentation (accepted)
rice360 conference for global health technologies · rice university