I had found reshape somewhat unintuitive, pivot_longer and separate_longer_delim from tidyr have been gamechangers for me

Understanding cross-reactivity is essential for TCR based therapies and we hope BATMAN as a stand alone tool, plus BATCAVE for future methods development can get us one step closer to this goal

We demonstrate the translational potential of BATMAN by accurately ranking immunogenic peptides of clinically relevant TCRs, essential for understanding
neoantigen immunogenicity and off-target effects of TCR-based therapies.

We then went on to develop on active learning framework for BATMAN, to minimize the number of TCR-pMHC experiments that need to be performed to learn the xreactivity of a novel TCR, thus guiding experimental biologists in the optimal strategy for designing experiments to determine TCR specificity.

BATCAVE became our benchmark to compare and develop BATMAN, a new probabilistic framework to predict TCR-pMHC activation and cross-reactivity. BATMAN learns interpretable features such as us AA-distance functions and epitope positional weights to correlate with biophysical TCR-pMHC properties

For BATCAVE, we scoured the TCR-pMHC literature from the past 20 years, collecting any mutational scan data that we could get our hands; we then processed these in a common framework, obtaining a databases of more than 22k+ TCR-pMHC activation datasets across human and mouse, CD4 and CD8 epitopes.

The work was a great collaborative effort between my lab and Saket Navlakha's lab at @cshlaboratory.bsky.social, largely driven by @amitava-immuno.bsky.social, @davipatti.bsky.social. Curious about BATCAVE and BATMAN? Read on ->