Author ORCID Identifier

https://orcid.org/0009-0003-9952-2377

Date of Award

Spring 6-3-2026

Document Type

Thesis (Undergraduate)

Department

Computer Science

First Advisor

Soroush Vosoughi

Second Advisor

Kenneth Hoehn

Abstract

Antibody heavy and light chain (H/L) pairing is fundamental to antigen recognition and stability. While single-cell sequencing preserves native pairing information, widely used bulk repertoire and spatial transcriptomics platforms do not, motivating the need for efficient ML methods to infer H/L pairing. Training a binary classifier for this task faces the methodological challenge of a lack of true biological negatives, since natural selection eliminates B cells with incompatible H/L pairs.

In this thesis, I introduce a biologically informed negative sampling strategy for H/L pairing classification, drawing on known V-gene biases in heavy and light chain pairing. Pseudo-negatives are constructed by sampling H/L sequences whose V-gene combinations are absent from the observed data. I train five classical ML models on amino acid composition features of H/L variable region sequences, and benchmark them against a state-of-the-art language model and a commonly used random shuffling strategy.

The results demonstrate that chain pairing can be predicted using relatively simple and interpretable classical approaches, with performance strongly dependent on biologically informed V-gene-based pseudo-negative sampling.

I contribute to two parallel efforts in computational immunology. First, the growing body of ML methods to pair, and generate paired, antibody data. Second, demonstrating the importance of negative sampling in a field where poor feature engineering, in the absence of true negatives, has led to reduced generalizability and overestimations of model performance.

Download

Share

COinS