Single-pass (bitopic) transmembrane (TM) proteins are highly abundant and functionally diverse, yet their structures remain poorly characterized. Experimental structures of full-length bitopic proteins and their functional complexes are scarce. Although thousands of predicted protein structures from hundreds of proteomes are available, accurately identifying bitopic proteins and determining their orientations within a lipid bilayer remain challenging. Here we developed AlphaPPM, a new pipeline and web server that identifies bitopic proteins from AlphaFold (AF) models and positions them in membranes. Built on the established Positioning of Proteins in Membranes (PPM) and Folding of Membrane-Associated Peptides (FMAP) approaches, AlphaPPM automatically removes low-confidence unstructured regions and signal sequences, separates structural domains, and then reassembles and orients extracellular, intracellular, and TM domains relative to membrane boundaries. We applied and validated the method across complete sets of bitopic proteins from 18 organisms. The results are available through MembranomeX, a dedicated web resource comprising an online database and four web servers. Supplemented by manual curation and cross-validation, the database contains membrane-oriented AF models of 16,697 bitopic proteins organized into 15 functional classes, 664 superfamilies, and 2497 families and data on 7966 protein complexes. Notably, the workflow identified 1225 bitopic proteins absent from established databases. MembranomeX integrates tools for structural modeling, data retrieval, model verification, and 3D visualization, providing a comprehensive platform for evolutionary, structural, and functional analyses of bitopic proteins.
Abstract T cell receptor (TCR) initiates T cell activation programs that are central to antiviral and antitumor immunity, and TCR-engineered T cell is a promising therapeutic approach in treating solid tumors. However, the design of TCR is time-consuming and labor-intensive, and computational TCR sequence generation remains comparatively underdeveloped with fewer dedicated models. We propose an encoder-decoder generation framework (i.e., Pep2Tcr) warm-started with pretrained peptide and TCR language models to conditionally generate TCR CDR3β sequences targeting a specific peptide. It also incorporates contrastive learning over batches to pull each TCR toward the peptide it pairs with. Through a series of progressive evaluations, Pep2Tcr outperforms other state-of-the-art methods under generation distribution, hit (independent testing sets), authenticity (third-party prediction), and generation probability (OLGA). Grafting and docking experiments of some predicted positive TCR sequences to the HLA-A*02:01-AAGIGILTV (melanoma antigen) complex showed plausible scores and potential interaction patterns for downstream experimental validation.
Abstract Reaction-conditioned enzyme recommendation ranks candidate enzymes for a query chemical transformation, but the contribution of explicit evolutionary context to this task remains insufficiently characterized. Here, we present an exploratory evaluation of multiple sequence alignment (MSA)-derived evolutionary features, integrating full-sequence and pocket-level MSA Transformer representations into EnzymeCage and a benchmark-adapted Horizyn implementation through concatenation or gated fusion. On the fixed Enzyme-405 candidate pools, the evaluated EnzymeCage Pocket-MSA model achieved a Top-1 success rate of 20.3% versus 14.6% for the baseline, while Full + Pocket MSA achieved a Top-10 success rate of 61.4% versus 55.6%; changes on Orphan-335 and in the adapted Horizyn implementation varied across configurations. Although a focused three-seed EnzymeCage baseline–Pocket comparison did not establish a consistent improvement across training runs and ranking cutoffs, analyses of MSA diversity, training-reference sequence similarity, and alignment-context removal characterize the observed effects and guide further evaluation of MSA-augmented enzyme recommendation.
Abstract Most CRISPR-Cas9 design tools select guide RNAs (gRNAs) by ranking predefined candidates rather than actively optimizing them. Here we instead formulate SpCas9 spacer refinement as a constrained search problem, in which a limited number of nucleotide substitutions may be introduced outside the PAM-proximal seed region. The search is driven by a thermodynamically grounded objective that combines CRISPRspec-based specificity with penalties for suboptimal GC content and long homopolymer runs. Within this single setting we compare five search strategies: three reinforcement-learning (RL) policies, which are Proximal Policy Optimization (PPO), Quantile-Regression Deep Q-Network (QR-DQN), and Rainbow Deep Q-Network (Rainbow DQN), together with a nonlearning score-guided search and a prompt-guided large language model (LLM). All five share the same objective, mutable positions, substitution budget, and evaluation protocol, so that only the search strategy differs. On 100 held-out IGF2R guides from a gene and chromosome excluded from training, every strategy significantly reduced the predicted off-target burden (38.6–78.3%) and improved independent specificity metrics without degrading on-target scores. The strategies nevertheless differed sharply in per-guide safety: the LLM achieved the largest aggregate reduction but introduced new off-target sites into 38% of guides that were initially free of them, whereas PPO, an on-policy RL algorithm, retained a strong gain with the lowest regression rate of 15%. These results show that the thermodynamic objective supplies a robust common optimization signal, while the choice of search strategy governs the trade-off between aggregate improvement and regression of already-favorable guides.
Abstract Substituents, decorating the central molecule core, play a key role in medicinal chemistry, where they are modified during lead optimization to improve biological activity, physicochemical properties, selectivity, and other drug properties. The selection of appropriate substituents, guided by the bioisosteric principle, is one of the most effective strategies for optimizing molecular properties while preserving affinity for the biological target. Despite the widespread application of bioisosteric replacement in drug discovery, systematic analyses of bioisosteric relationships among substituents derived from large collections of bioactive molecules remain limited. In this work, we present a comprehensive analysis of substituents occurring in bioactive molecules and investigate their bioisosteric relationships using data extracted from medicinal chemistry literature. Bioisosteric pairs are identified from substitution patterns observed in related compounds with activity on the same target, enabling the construction of a network of frequently interchangeable substituents. Because the analysis is based on the observed co-occurrence of substituents within the same molecular environments rather than on structural similarity or physicochemical descriptors, it captures experimentally validated substitution patterns and uncovers not only well-established bioisosteric pairs but also less well-known relationships. The resulting bioisosteric substituent network is made available through a free web application at https://ertlmolecular.com/tools.html, providing a practical resource for medicinal chemists engaged in lead optimization, analog design, and library development.
