Arman Bekov

Sanjar Khudoyberdiev, Arman Bekov

Neural models for TCR-pMHC binding prediction are susceptible to shortcut learning: they exploit spurious correlations in training data -- such as peptide length bias or V-gene co-occurrence -- rather than the physical binding interface. This renders predictions brittle under family-held-out and distance-aware evaluation, where such shortcuts do not transfer. We introduce \emph{Counterfactual Invariant Prediction} (CIP), a training framework that generates biologically constrained counterfactual peptide edits and enforces invariance to edits at non-anchor positions while amplifying sensitivity at MHC anchor residues. CIP augments the base classifier with two auxiliary objectives: (1) an invariance loss penalizing prediction changes under conservative non-anchor substitutions, and (2) a contrastive loss encouraging large prediction changes under anchor-position disruptions. Evaluated on a curated VDJdb-IEDB benchmark under family-held-out, distance-aware, and random splits, CIP achieves AUROC 0.831 and counterfactual consistency (CFC) 0.724 under the challenging family-held-out protocol -- a 39.7\% reduction in shortcut index relative to the unconstrained baseline. Ablations confirm that anchor-aware edit generation is the dominant driver of OOD gains, providing a practical recipe for causally-grounded TCR specificity modeling.

Arman Bekov, Timur Bekzhanov, Bekzat Sadykov

Predicting T-cell receptor (TCR)--peptide-MHC (pMHC) binding is central to vaccine design and T-cell therapy, yet deployed models frequently encounter epitopes unseen during training, causing silent overconfidence and unreliable prioritization. We address this by framing TCR--pMHC prediction as a \emph{selective prediction} problem: a calibrated model should either output a trustworthy confidence score or explicitly abstain. Concretely, we (1) introduce a dual-encoder architecture encoding both CDR3$α$/CDR3$β$ and peptide sequences via a pre-trained protein language model; (2) apply temperature scaling to correct systematic probability miscalibration; and (3) impose a conformal abstention rule that provides finite-sample coverage guarantees at a user-specified target error rate. Evaluated under three split strategies -- random, epitope-held-out, and distance-aware -- our method achieves AUROC 0.813 and ECE 0.043 under the challenging epitope-held-out protocol, reducing ECE by 69.7\% relative to an uncalibrated baseline. At 80\% coverage, the selective model further reduces error rate from 18.7\% to 10.9\%, demonstrating that calibrated abstention enables principled coverage-risk trade-offs aligned with practical screening budgets.

Arman Bekov, Azamat Nurgali

Online education platforms have experienced explosive growth over the past decade, generating massive volumes of user-generated content in the form of reviews, ratings, and behavioral logs. These heterogeneous signals provide unprecedented opportunities for understanding learner satisfaction, which is a critical determinant of course retention, engagement, and long-term learning outcomes. However, accurately predicting satisfaction remains challenging due to the short length, noise, contextual dependency, and multi-dimensional nature of online reviews. In this paper, we propose a unified \textbf{Large Language Model (LLM)-based multi-modal framework} for predicting both platform-level and course-level learner satisfaction. The proposed framework integrates three complementary information sources: (1) short-text topic distributions that capture latent thematic structures, (2) contextualized sentiment representations learned from pretrained Transformer-based language models, and (3) behavioral interaction features derived from learner activity logs. These heterogeneous representations are fused within a hybrid regression architecture to produce accurate satisfaction predictions. We conduct extensive experiments on large-scale MOOC review datasets collected from multiple public platforms. The experimental results demonstrate that the proposed LLM-based multi-modal framework consistently outperforms traditional text-only models, shallow sentiment baselines, and single-modality regression approaches. Comprehensive ablation studies further validate the necessity of jointly modeling topic semantics, deep sentiment representations, and behavioral analytics. Our findings highlight the critical role of large-scale contextual language representations in advancing learning analytics and provide actionable insights for platform design, course improvement, and personalized recommendation.