How dhar7/compbioproject works
This repository implements a reproducible TCR–peptide–MHC negative-set research study. Its workflow prepares positive-pair and healthy-repertoire inputs, curates them into peptide-grouped folds, then runs model/control experiments and reports discrimination, chance thresholds, and uncertainty. Shared study utilities provide sampling, sequence encoding, training, and metrics. The tree also includes epiTCR and TITAN prediction-score utilities. Source excerpts show the main preparation and analysis calls; predictor replication details and some shared helpers are not shown, so those relationships are not inferred.
Main entry points: argha.py Control experiments · farhan.py Curate and fold · prepare_data.py Prepare inputs · data Positive and repertoire tables
Components 15
Input preparation
- Study datasets
data - Prepare inputs
prepare_data.py - Positive and repertoire tables
data
Study curation
- Curate and fold
farhan.py - Processed folds
data/processed - Shared study methods
study_common.py
Model analysis
- Control experiments
argha.py - Frozen ESM-2 encoder
study_common.py - Negative-set sampling
study_common.py - Control model training
study_common.py - Metrics and uncertainty
study_common.py - Study result tables
outputs/argha
Predictor scoring
- epiTCR score utility
tcr_predictors/epitcr/score.py - TITAN score utility
tcr_predictors/titan/score.py
Other components
- Prediction files
Connections 14
- Study datasets → Prepare inputs: provides source files
prepare_data.py - Prepare inputs → Positive and repertoire tables: writes tables
prepare_data.py - Positive and repertoire tables → Curate and fold: loads inputs
farhan.py - Curate and fold → Shared study methods: uses helpers
farhan.py - Curate and fold → Processed folds: writes folds
farhan.py - Processed folds → Control experiments: loads processed data
argha.py - Control experiments → Shared study methods: uses helpers
argha.py - Control experiments → Frozen ESM-2 encoder: constructs encoder
argha.py - Control experiments → Negative-set sampling: samples labeled splits
argha.py - Control experiments → Control model training: fits controls
argha.py - Control experiments → Metrics and uncertainty: computes metrics
argha.py - Control experiments → Study result tables: writes summaries
argha.py - Prediction files → epiTCR score utility: reads predictions
tcr_predictors/epitcr/score.py - Prediction files → TITAN score utility: reads predictions
tcr_predictors/titan/score.py