MOTIVATION: Reliable TCR-epitope binding prediction models are essential for development of adoptive T cell therapy and vaccine design. These models often struggle with false positives, which can be attributed to the limited data coverage in existing negative sample datasets. Common strategies for generating negative samples, such as pairing with background T cell receptors (TCRs) or shuffling within pairs, fail to account for model-specific vulnerabilities or biologically implausible sequences. RESULTS: To address these challenges, we propose an iterative attack-and-defend framework that systematically identifies and mitigates weaknesses in TCR-epitope prediction models. During the attack phase, a reinforcement learning from AI feedback (RLAIF) framework is used to attack a prediction model by generating biologically implausible sequences that can easily deceive the model. During the defense phase, these identified false positives are incorporated into fine-tuning dataset, enhancing the model's ability to detect false positives. A comprehensive adversarial negative dataset can be obtained by iteratively attacking and defending the model. This dataset can be directly used to improve model robustness, eliminating the need for users to conduct their own attack-and-defend cycles. We apply our framework to five existing binding prediction models, spanning diverse architectures and embedding strategies to show its efficacy. Experimental results show that our approach significantly improves these models' ability to detect adversarial false positives. The combined dataset constructed from these experiments also provides a benchmarking tool to evaluate and refine prediction models. Our framework offers a new approach for improving model robustness in other biological tasks where negative sampling is inherently limited. AVAILABILITY AND IMPLEMENTATION: The curated dataset and code are available at a public repository (https://github.com/Lee-CBG/BAP_Attack_n_Defend).

T cell receptors (TCRs) are critical components of adaptive immune systems, responsible for responding to threats by recognizing epitope sequences presented on host cell surface. Computational prediction of binding affinity between TCRs and epitope sequences using machine/deep learning has attracted intense attention recently. However, its success is hindered by the lack of large collections of annotated TCR-epitope pairs. Annotating their binding affinity requires expensive and time-consuming wet-lab evaluation. To reduce annotation cost, we present ActiveTCR, a framework that incorporates active learning and TCR-epitope binding affinity prediction models. Starting with a small set of labeled training pairs, ActiveTCR iteratively searches for unlabeled TCR-epitope pairs that are "worthy" for annotation. It aims to maximize performance gains while minimizing the cost of annotation. We compared four query strategies with a random sampling baseline and demonstrated that ActiveTCR reduces annotation costs by approximately 40%. Furthermore, we showed that providing ground truth labels of TCR-epitope pairs to query strategies can help identify and reduce more than 40% redundancy among already annotated pairs without compromising model performance, enabling users to train equally powerful prediction models with less training data. Our work is the first systematic investigation of data optimization for TCR-epitope binding affinity prediction.

Interpretable machine learning has gained much attention recently. Briefness and comprehensiveness are necessary in order to provide a large amount of information concisely when explaining a black-box decision system. However, existing interpretable machine learning methods fail to consider briefness and comprehensiveness simultaneously, leading to redundant explanations. We propose the variational information bottleneck for interpretation, VIBI, a system-agnostic interpretable method that provides a brief but comprehensive explanation. VIBI adopts an information theoretic principle, information bottleneck principle, as a criterion for finding such explanations. For each instance, VIBI selects key features that are maximally compressed about an input (briefness), and informative about a decision made by a black-box system on that input (comprehensive). We evaluate VIBI on three datasets and compare with state-of-the-art interpretable machine learning methods in terms of both interpretability and fidelity evaluated by human and quantitative metrics.

MOTIVATION: Gene clusters are arrangements of functionally related genes on a chromosome. In bacteria, it is expected that evolutionary pressures would conserve these arrangements due to the functional advantages they provide. Visualization of conserved gene clusters across multiple genomes provides key insights into their evolutionary histories. Therefore, a software tool that enables visualization and functional analyses of gene clusters would be a great asset to the biological research community. RESULTS: We have developed GeneclusterViz, a Java-based tool that allows for the visualization, exploration and downstream analyses of conserved gene clusters across multiple genomes. GeneclusterViz combines an easy-to-use exploration interface for gene clusters with a host of other analysis features such as multiple sequence alignments, phylogenetic analyses and integration with the KEGG pathway database. AVAILABILITY: http://biohealth.snu.ac.kr/GeneclusterViz/; http://microbial.informatics.indiana.edu/GeneclusterViz/

Discordant read pairs [ 1 , 2 ] – those deviating either from expected insert size range or correct relative orientation – have served as vital clues to identifying structural variants (SV) in genomes. Collecting discordant read pairs is the first step in SV detection and is often done by sequence alignment. When there are repetitive elements, such as insertion sequence (IS), a class of transposable elements in bacterial genomes, discordant read pairs can have multiple mapping loci – making them more challenging to be placed and interpreted. Instead of resolving such tangled mapping results, many tools simply ignore these mapped read pairs, potentially missing SVs involving repetitive elements. We present an idea of using approximate de Bruijn graphs ( A -Bruijn graphs) [ 3 ] to identify discordant read pairs, in order to discover SVs. Repeats are easily recognized in A -Bruijn graphs, as all repetitive elements of the same kind are collapsed into a contiguous edge. When read pairs representing repetitive elements are mapped to a reference A -Bruijn graph, only those from novel insertions are flagged as discordant and the rest – those from preexisting insertion loci – mapped concordantly. We applied this approach to whole genome sequencing data [ 4 ] (~100x per sample using 90bp x 2 paired end Illumina sequencing) obtained from 38 lines of Escherichia coli PFM2, a derivative strain of E. coli K-12 MG1655, and 34 lines of a mismatch repair deficient (deletion of mutL ) derivative that were propagated for ~3,080 and ~375 generations respectively via a mutation accumulation (MA) strategy. All of the inferred IS insertions were directly confirmed by PCR experiments. A total of 27 IS transpositions has been detected and includes 5 out of 12 IS families present in E. coli K-12. We have also identified an insertion of IS186 that is fixed among all MA lines and not present in the reference E. coli genome. 24 out 27 inferred insertions were validated by PCR and 3 of them are currently under analysis. The fixed insertion of IS186 in the samples was also confirmed by PCR. Our method can pinpoint SVs by identifying discordant read pairs resulting from novel insertions of repetitive elements, where many other currently available tools fail. This result serves as a first step towards inferring the neutral rate of IS transposition in bacterial genomes.