Aligning molecular sequence representations (e.g., SMILES notations) with textual descriptions is critical for applications spanning drug discovery, materials design, and automated chemical literature analysis. Existing methodologies typically treat molecular captioning (molecule-to-text) and text-based molecular design (text-to-molecule) as separate tasks, relying on supervised fine-tuning or contrastive learning pipelines. These approaches face three key limitations: (i) conventional metrics like BLEU prioritize linguistic fluency over chemical accuracy, (ii) training datasets frequently contain chemically ambiguous narratives with incomplete specifications, and (iii) independent optimization of generation directions leads to bidirectional inconsistency. To address these issues, we propose RTMol, a bidirectional alignment framework that unifies molecular captioning and text-to-SMILES generation through self-supervised round-trip learning. The framework introduces novel round-trip evaluation metrics and enables unsupervised training for molecular captioning without requiring paired molecule-text corpora. Experiments demonstrate that RTMol enhances bidirectional alignment performance by up to 47% across various LLMs, establishing an effective paradigm for joint molecule-text understanding and generation.

MOTIVATION: T cell receptor (TCR) recognition of peptide-major histocompatibility complex (pMHC) complexes is central to adaptive immunity, yet rational design of immunogenic epitopes remains elusive due to complex triplet binding constraints and data scarcity. No existing method can generate epitopes satisfying simultaneous requirements for antigenicity, MHC presentation, and TCR specificity. RESULTS: We present EPIC, a multi-objective diffusion framework that decomposes TCR-pMHC binding into three biologically grounded sub-tasks, enabling training-free gradient guidance without end-to-end retraining. By integrating ESM-based classifiers with a peptide diffusion generator, EPIC leverages heterogeneous immunological interaction datasets to generate diverse, context-aware epitopes. EPIC-designed top-three epitopes achieve lower predicted interface energies compared to ground-truth epitopes in 78.31% of test cases, while maintaining 80.1% sequence novelty and comparable structural confidence. Generated epitopes exhibit 100% uniqueness, high diversity (64.05%), and high antigenicity scores (0.4723). To our knowledge, EPIC is the first computational framework capable of de novo epitope design while explicitly integrating the triplet constraints of TCR-pMHC binding. This paradigm shift from discovery to design unlocks new potential for personalized cancer vaccines, precision adoptive T cell therapy, and rapid response to emerging infectious diseases. AVAILABILITY AND IMPLEMENTATION: The source code of EPIC is available at https://github.com/Octopus125/EPIC and archived on Zenodo (DOI: 10.5281/zenodo.18537646).

Chemical reaction prediction remains a fundamental challenge in organic chemistry, where existing machine learning models face two critical limitations: sensitivity to input permutations (molecule/atom orderings) and inadequate modeling of substructural interactions governing reactivity. These shortcomings lead to inconsistent predictions and poor generalization to real-world scenarios. To address these challenges, we propose ReaDISH, a novel reaction prediction model that learns permutation-invariant representations while incorporating interaction-aware features. It introduces two innovations: (1) symmetric difference shingle encoding, which extends the differential reaction fingerprint (DRFP) by representing shingles as continuous high-dimensional embeddings, capturing structural changes while eliminating order sensitivity; and (2) geometry-structure interaction attention, a mechanism that models intra- and inter-molecular interactions at the shingle level. Extensive experiments demonstrate that ReaDISH improves reaction prediction performance across diverse benchmarks. It shows enhanced robustness with an average improvement of 8.76\% on R$^2$ under permutation perturbations.

Despite the considerable progress that has been made in cancer research, identifying cancer genes remains a significant challenge due to the intricate nature of the disease. Given the importance of incorporating gene interaction relationships in identifying potential cancer genes, Graph Neural Networks (GNNs) have garnered increasing attention for their ability to model gene associations effectively. Particularly, recent studies have demonstrated the superiority of GNNs in deciphering the knowledge embedded in Protein-Protein Interaction (PPI) networks for cancer gene prediction. However, these studies primarily focused on a single PPI network, overlooking the valuable insights encapsulated within other PPI networks. Additionally, previous endeavors often centered around pan-cancer datasets, neglecting the importance of predicting specific cancer genes. To address these limitations, we present a novel method called MPIT, which employs Graph Transformer Networks (GTNs) to identify specific cancer driver genes. MPIT effectively integrates data from diverse PPI and multi-omics data via the alignment and fusion of gene representations learned from different PPI networks. We collect three distinct cancer cell line datasets to assess the model performance. Our experimental findings demonstrate the superiority of MPIT over the existing methods, achieving the state-of-the-art performance across all three datasets.

Recent breakthrough in spatial transcriptomics has brought great opportunities for exploring gene regulatory networks (GRNs) from a brand-new perspective. Especially, the local expression patterns and spatio-temporal regulation mechanisms captured by spatial expression images allow more delicate delineation of the interplay between transcript factors and their target genes. However, the complexity and size of spatial image collections pose significant challenges to GRN inference using image-based methods. Extracting regulatory information from expression images is difficult due to the lack of supervision and the multi-instance nature of the problem, where a gene often corresponds to multiple images captured from different views. While graph models, particularly graph neural networks, have emerged as a promising method for leveraging underlying structure information from known GRNs, incorporating expression images into graphs is not straightforward. To address these challenges, we propose a two-stage approach, MIGGRI, for capturing comprehensive regulatory patterns from image collections for each gene and known interactions. Our approach involves a multi-instance graph neural network (GNN) model for GRN inference, which first extracts gene regulatory features from spatial expression images via contrastive learning, and then feeds them to a multi-instance GNN for semi-supervised learning. We apply our approach to a large set of Drosophila embryonic spatial gene expression images. MIGGRI achieves outstanding performance in the inference of GRNs for early eye development and mesoderm development of Drosophila, and shows robustness in the scenarios of missing image information. Additionally, we perform interpretable analysis on image reconstruction and functional subgraphs that may reveal potential pathways or coordinate regulations. By leveraging the power of graph neural networks and the information contained in spatial expression images, our approach has the potential to advance our understanding of gene regulation in complex biological systems.