Briefings in Bioinformatics

Discovery of pathway associations and druggability can leverage underutililized dark kinase genes for treating complex diseases (proven for cancer and neurodegeneration), boosted with computational methods. Herein, we employ BERT-based embeddings of proteins and pathways (refined via two-stage transformer and heterogeneous graph transformer) and protein-protein and protein-pathway associations-both positive (curated from databases) and negative (generated using Wasserstein Generative Adversarial Networks with gradient penalty) to train XGBoost and lightGBM classifiers for predicting pathways associated to human dark kinase proteins, with important features unveiled through SHAP analysis. All pathways are clustered and proteins related to same pathway clusters are grouped together (via predicted and positive protein-pathway associations). Selected PCOS-related human dark kinase proteins (with high predicted and existent associations to PCOS pathways) are docked with known PCOS drugs for druggability analysis. Our model attains accuracy, F1-score, specificity, MCC, AUROC and AUPRC of 0.9816, 0.9816, 0.9852, 0.9632, 0.9978 and 0.9982 respectively, supersedes existing work, correctly classifies 97.48% of test data, predicts 62225 pathway associations to above proteins, infers functional similarity of 96 such proteins to human protein(s) and traces nine important positive features. Our model can be used to determine varied functionalities and disease relevance of proteins via predicted pathway associations.

Evaluating protein sequence similarity remains challenging in the protein-sequence twilight zone (20-35% sequence identity), where traditional methods often fail. In this study, we evaluate whether mean-pooled embeddings from four protein language models: ESM-1b, ESM-2, ProtT5, and ProstT5 can estimate pairwise structural similarity without performing sequence alignment. The benchmark dataset includes 20,445 PISCES protein pairs with sequence identity [≤]30%, representing the protein-sequence twilight zone, with TM-align-derived TMmin used as the structural ground truth. Protein embeddings are compared using cosine similarity, Euclidean- and Manhattan-derived similarities, an RBF kernel, and dot product. Among these similarity metrics, cosine similarity performs best across all four models. Moreover, ProstT5 achieves the highest Spearman correlation with TMmin, followed by ESM-2, ProtT5, and ESM-1b, while all four PLMs outperform BLASTP overall. Furthermore, the advantage of PLM embeddings is most pronounced for protein pairs with the lowest sequence identity. ProstT5 also provides the best discrimination between structurally similar and dissimilar protein pairs. Moreover, it offers a favorable balance between similarity performance and the computational requirements of residue-level embedding generation and storage. Overall, these findings support PLM embeddings as an effective alignment-free approach for detecting structural relationships among proteins in the twilight zone.

Deep learning methods, such as AlphaFold and RosettaFold, achieve high accuracy in protein structure prediction. However, predicting the structure of large protein complexes remains challenging due to their large size and intricate multi-chain interactions. Docking-based methods can handle large proteins, but are limited by the huge combinatorial binding space of multichains. Assembly-based approaches offer an alternative, but their accuracy critically relies on the precision of predicted subcomponents. Addressing the challenges, we propose HDOCK-Multimer (HDM), a structure prediction framework of large protein complexes by integrating ab initio docking and combinatorial assembly. HDM can efficiently reduce reliance on subcom-ponent accuracy through docking process, while leveraging the pairwise interactions of subcom-ponents through assembly strategy. HDM is extensively validated on three benchmarks of 35 large heteromeric complexes, 172 large protein complexes, and 7 CASP15 targets, and compared with state-of-the-art methods including MoLPC, CombFold, AlphaFold-Multimer (AFM), and AlphaFold3 (AF3). It is shown that HDOCK-Multimer substantially outperforms the other methods. In addition, HDM also shows ability to predict the stoichiometry and model the complex without stoichiometry input. It is anticipated that HDM will serve as a powerful tool for studying large protein complexes or molecular machines. The HDM package is freely available at https://github.com/huang-laboratory/HDOCK-Multimer/.

Abstract Motivation: Antibodies are a major class of therapeutic molecules, and their recognition of target antigens is largely mediated by complementarity-determining regions (CDRs), making antigen-conditioned CDR design a central problem in antibody engineering. Recent generative methods have enabled antigen-conditioned co-design of CDR sequences and structures, but their limited capacity to capture local interface heterogeneity and lack explicit energy-based guidance during sampling, which may result in unfavorable antibody-antigen interaction energies. Overcoming these limitations requires methods that better represent diverse interface environments via adaptive routing and incorporate physical guidance to steer sampling toward energetically favorable conformations. Results: We present ForceFlowAb, a physics-aware mixture-of-experts flow-matching framework for antigen-conditioned CDR sequence-structure co-design. The framework models heterogeneous interface environments through specialized expert routing and applies differentiable force-field guidance during sampling to guide CDR generation toward energetically favorable conformations. For CDR-H3 design, ForceFlowAb achieved more favorable antibody-antigen interaction energies than FlowDesign and Diffab, with improvement rates (IMP) of 46.5% versus 35.0% and 35.5%, respectively. For simultaneous six-CDR design, ForceFlowAb also outperformed Diffab, with IMP values of 16% versus 9%. These results suggest complementary roles for interface-adaptive modeling and energy-based guidance, with the former capturing binding-mode diversity and the latter leveraging physical constraints to ensure biophysical feasibility. Availability and implementation: The web server is freely available at http://zhanglab-bioinf.com/ForceFlowAb. The source code and implementation are available at https://github.com/iobio-zjut/ForceFlowAb. Contact: zgj@zjut.edu.cn Supplementary information: Supplementary data are available at Bioinformatics online.

Motivation: Traditional methods primarily rely on statistical features such as k-mers and GC content, making it difficult to capture complex internal relationships within sequences. Deep learning models typically rely on discrete encodings, leading to issues such as information sparsity, dimensional redun-dancy, and disruption of sequence continuity. Our previous UniWave-1 framework transforms se-quences into waveform signals; however, it lacks comprehensive physicochemical integration and com-putational efficiency. Results: We propose the UniWave-2 feature extraction framework. UniWave-2 incorporates three key advancements: (i) integration of hydrophobicity into the encoding scheme; (ii) enhancement of wave-form resolution through Fourier transform, coupled with wavelet transform to precisely capture both local details and global periodic patterns; and (iii) development of a lightweight multi-scale GRU model that facilitates cross-dimensional feature interactions, captures bidirectional temporal dependencies, and integrates long- and short-range patterns with spatial positional awareness, thereby yielding rich feature representations. UniWave-2 achieved competitive performance across multiple genomic tasks, while ISM analysis demonstrated that waveform encoding enables interpretable attribution without re-lying on gradient information. These results establish UniWave-2 as a lightweight and interpretable framework for nucleic acid sequence analysis.

Pediatric low-grade glioma (pLGG) is the most common type of brain tumor in children, accounting for approximately 30% of all central nervous system tumors in children. pLGG has multiple molecular subtypes that differ in disease progression, recurrence patterns, and treatment responses. Conventional wet lab approaches including molecular profiling and histopathological studies for pLGG characterization are time consuming, costly, and laborious. Recently, methods based on artificial intelligence (AI) or machine learning (ML) have been widely used for pLGG molecular categorization, but most of them can only identify two or three pLGG subtypes. To more comprehensively characterize the molecular subtypes of pLGG and their potential biological and therapeutic significance, we develop an integrated meta-clustering approach, namely Meta-pLGG, that can explore high resolution molecular subtypes and their transcriptional heterogeneity for pLGG. Specifically, we first performed multiple rounds of random projection (RP) to generate dimension-reduced feature vectors from pLGG transcriptomics data, each of which was subsequently clustered by different clustering algorithms including hierarchical clustering, K-means, Self-Organizing Maps (SOM), Non-negative Matrix Factorization (NMF), Gaussian Mixture Model (GMM), and Spectral Clustering, as base clustering methods. Then, to yield robust clustering performance, we integrated the clustering results of these RP based individual clustering algorithms by adopting a weighted meta-clustering (wMetaC) approach. Results based on 532 pLGG patients suggested that our proposed approach demonstrated superior stability and discriminative powers for higher resolution pLGG subtyping compared to conventional approaches. Based on consensus matrix analysis, we identified two major pLGG mega-subtypes, with one further subdivided into three subgroups and the other into two. Then, we performed cluster specific differential gene expression analysis, molecular pathway analysis, and gene-drug-disease association analysis. The results showed that the identified five subgroups exhibited significant subtype-specific transcriptomic heterogeneity. In summary, our meta-clustering approach demonstrated much higher performance and robustness in identifying higher resolution molecular subtypes of pLGG, revealing the molecular heterogeneity within pLGG and potentially providing new insights for more precise molecular subtyping and precision therapy.

Accurately predicting the effects of mutations on protein-RNA binding is crucial for elucidating disease mechanisms. Yet, exhaustively exploring the space of all possible variants is prohibitively expensive, motivating computational methods that can quantify mutation-induced changes in binding affinity (aka {Delta}{Delta}G) accurately and efficiently. We present iSCALE, an interpretable and generalizable deep learning method that adopts an implicit Spatial Coupling-Aware Ligand Encoding strategy to predict mutation-induced binding affinity changes. By injecting this implicit multiscale encoding scheme into a bidirectional state space modeling architecture, iSCALE learns a generalizable multiscale coupling pattern that achieves superior performances on not only the protein-RNA binding {Delta}{Delta}G, but also the protein stability {Delta}{Delta}G and protein-protein binding {Delta}{Delta}G predictions. Detailed analyses demonstrate that the model attention scores align well with structural characteristics. In addition, iSCALE shows good discriminative ability when predicting close samples such as complexes of same mutation but with different ligands or the same complex but with different mutation sites. In summary, iSCALE serves as an effective in silico tool for large-scale protein-RNA binding {Delta}{Delta}G prediction, which pushes the border of understanding in mutation-induced pathological outcomes.

BackgroundHighly variable gene (HVG) selection begins almost every single-cell RNA-seq analysis. While ranking formulas have been compared extensively, the integer gene budget at which any ranking must be truncated is typically left to the user and habitually fixed near 2,000. Relying on such a convention carries hidden costs--lists that are too short erase subtle structure, whereas lists that are too long add noise and computational overhead. Moreover, because global rankings measure variance across all cells, markers for rare populations often lose the "variance vote count" to dominant bulk variation, leading to an unfair feature allocation at the hard cutoff. Whether this convention is defensible, and whether the budget and tail can be set from data without disturbing the ranking, has not been examined systematically. ResultsUnder a frozen seurat_v3 ranking, k-sweeps across 18 labeled datasets show that n = 2,000 is ARI-optimal on 1 of 18 datasets and that the best available budget is worth a mean ARI gain of +0.033 over it, establishing cardinality as a real and largely unexploited design axis. We present scFair, a Scanpy-compatible HVG layer that automates list length alone: geometry-aware auto_n sets a base size k from multi-seed density and stability features of an intermediate embedding (trading a modest, intentional compute increase for a safer data-driven default), and a same-rank append step acts as a conservative safeguard against cutoff unfairness by adding a short near-miss tail. The ranking is never recomputed or reweighted. On the 18-dataset panel, the default path improved Leiden-label agreement over HVG@2000 (median {Delta}ARI = +0.016; 13/5; Wilcoxon P = 0.0077) and outperformed the neighborhood-based selector triku at author defaults on 15/18 datasets (median +0.024; P = 0.004), while triku did not improve on HVG@2000. Controls locate the effect: cell-number-only rules do not beat HVG@2000, an FDR-chosen length imposed on the frozen ranking is flat, and a fixed HVG@2200 default is not a general substitute because it cannot produce the short lists that compact matrices call for. ConclusionsA fixed budget near 2,000 HVGs is frequently suboptimal, and list cardinality is a separable design axis that can be automated without changing the ranking formula. Effect sizes are modest, the short-list branch rests on four datasets, and rule thresholds were developed with partial overlap to the evaluation panel.

Large-scale prediction and assessment of clinical patient responses (i.e., RECIST class) to drug combinations remains challenging due to scarce patient-derived data. The existing prediction methods mainly rely on cancer cell line models. However, substantial biological heterogeneity between cancer cell lines and cancer patients within same tissues, as well as the heterogeneity between one tissue and another, often limit the generalizability of these methods in clinical patients. To overcome these limitations, here we present CaMeRe, a Causally-inspired Meta-representation learning framework designed to predict patient-specific clinical Response to drug combinations. In situations where stable causal factors and domain-specific response-modulating factors are unobservable, explicit discrete domain labels are unavailable, and data is scarce, CaMeRe designed a domain-invariant causal representation learning (DICRL) model guided by the invariant information bottleneck theory and causal intervention invariance principle, and also built a meta-learning framework with bi-level domain generalization to optimize DICRL model for achieving multi-domain generalization within and across tissues. By integrating the causal representation learning and meta learning framework, CaMeRe not only exhibited robust multi-domain generalization performance across multiple clinical drug combination response datasets and PDXs drug combination response datasets and generalization scenarios, but also had better interpretability. We applied CaMeRe to predict drug-combination response scores for 3,423 patients across 542,080 drug combinations. The predicted scores were significantly associated with biomarkers of known drug combinations and enabled the prioritization of candidate drug combinations across 11 cancer types, with stronger support from literature and clinical trial evidences than random baselines. We believe that CaMeRe can be a useful tool for predicting large-scale clinical individual drug combination responses and it has broad clinical applications.

Single-cell foundation models (scFMs) are commonly benchmarked using final-layer embeddings, although useful depth may depend on the task. We evaluate Geneformer, scBERT, scGPT, and scFoundation as frozen feature extractors for cell-type and perturbation-response prediction using layer-wise probes, leave-one-dataset-out (LODO) selection, readout controls, representation diagnostics, centered kernel alignment (CKA), and retrieval/collapse analyses. Intermediate layers significantly improve cell-type accuracy for scGPT and scBERT in several settings, while Geneformer and scFoundation generally show performance comparable to their final layers. For perturbation prediction, LODO identifies layer 1 as an earliest-on-par candidate across all four models, without establishing universal superiority. Paired scGPT CLS and mean-pooling analyses show that readout affects preferred depth. CKA and representation metrics reveal model- and task-dependent geometric changes, while retrieval and collapse checks support the robustness of perturbation findings. We recommend reporting extraction depth, readout, selection protocol, and task-specific diagnostics. A validated non-final layer may also reduce computation, although efficiency was not directly benchmarked.

Accurate prediction of peptide-MHC class II (pMHC-II) binding remains challenging because of extensive MHC polymorphism and context-dependent peptide recognition, and it is unclear which protein representation best captures these determinants when the downstream model is held fixed. Here, we present PREpiBind, a pMHC-II prediction method based on a dual-stream, joint-attention framework that integrates MHC and epitope representations. We integrated ten protein representations spanning substitution-matrix, structure-prediction-derived, and protein language model (PLM) families and evaluated them under a common downstream architecture and identical splits across Qualitative, mass spectrometry (MS), and thresholded IC50 datasets. Overall, PREpiBind with PLM representations yielded higher performance than the evaluated reference methods in the Qualitative and MS datasets, whereas NetMHCIIpan-4.3 was higher on the thresholded IC50 datasets using its binding-affinity head. In pooled evaluations, PLM representations yielded the highest ROC-AUC values among the tested representations. ESM3 Large led the Qualitative dataset with an AUC value of 0.927 +/- 0.002, and PLMs also led on the MS and thresholded IC50 datasets. This advantage remained, but narrowed in allele-wise and leave-one-molecule-out evaluations, where structure-prediction-derived Chai-1 was competitive with the leading PLM representations. Under cross-species H2-out transfer, PLMs led when all H2 rows were pooled, whereas Chai-1 led when the eight H2 molecules were weighted equally. The small H2 panel did not support a stable ordering among these leading representations. The results indicate that protein representation choice should depend on the intended prediction scenario rather than on a single global ranking. PREpiBind is an openly available pMHC-II prediction framework with modular and flexible protein representations.

Single-cell foundation models have recently emerged as a promising approach for learning general- purpose representations from large-scale transcriptomic data. These models are trained on millions of cells and are designed to transfer their learned representations to a wide range of downstream tasks. However, their practical benefits compared to traditional approaches are still not fully understood. This study evaluates four foundation models, namely scGPT, SCimilarity, UCE, and Transcriptformer, across four downstream tasks: cell type annotation, human data integration, cross-species data integration, and protein expression prediction. Embeddings generated by each model were assessed using multiple public single-cell datasets and compared against conventional machine learning baselines. Performance was measured using task-specific evaluation metrics, including classification, integration, and regression metrics. The results showed that foundation model embeddings did not consistently outperform traditional approaches. In the cell type annotation task, baseline methods achieved the strongest performance across most datasets. For protein expression prediction, however, embeddings from the foundation models generally produced more accurate predictions than the baseline, with SCimilarity achieving the lowest prediction error and Transcriptformer obtaining the highest correlation scores. In the data integration task, all foundation models produced moderate results, while scVI (the baseline) achieved the strongest integration performance. Overall, the results suggest that current single-cell foundation models provide useful representations for some downstream tasks in zero-shot conditions but do not yet offer a universal replacement for task-specific methods. Their effectiveness remains dependent on the application and evaluation setting.

Many computational methods to predict intrinsic disordered regions (IDRs) in proteins have been developed and their performances are blindly evaluated in community-driven assessment, Critical Assessment of protein Intrinsic Disorder (CAID). In this study, we developed PredIDR3 series, an updated version of PredIDR2 tested in CAID3 to accurately predict IDRs from protein sequences. It includes two methods depending on ensemble way. The performances of PredIDR3 series (AUC_ROC=0.953) are remarkably better than our previous PredIDR2 (AUC_ROC=0.936) on Disorder-PDB dataset of CAID3, which is thought to be mainly attributed to the use of more information for intrinsic disorder prediction based on deep convolutional neural network. In details, we introduced a new output-encoding scheme permitting a large sliding window (size=91) for the first time and extracted negative samples of the training set from non-IDRs of both PDB and DisProt databases, allowing to use more information for prediction of intrinsic disorder. PredIDR3 achieved comparable performance to the top-ranking methods of CAID3 in all criteria measured. PredIDR3 series can be freely available through the CAID Prediction Portal at https://caid.idpcentral.org/portal or downloaded as a Singularity container from https://biocomputingup.it/shared/caid-predictors/.

Recent advances in artificial intelligence have accelerated the discovery of bioactive peptides by enabling computational exploration of the vast peptide sequence space. However, existing peptide generation approaches generally rely on either distribution-learning models, which generate biologically realistic sequences but do not consistently optimize functional activity, or optimization-based methods, which maximize prediction confidence while often deviating from the underlying distribution of experimentally validated peptides. To address this limitation, a two-phase generative-evolutionary framework is proposed that integrates distribution learning with evolutionary optimization. In the first phase, Variational Autoencoders (VAE), Autoregressive Transformers (ART), and Token Diffusion Transformers (TDT) are used to generate biologically plausible seed peptides. In the second phase, these peptides were used as initial seed for Hill Climbing optimization procedure that iteratively improves fitness function score. The proposed two-phase framework was evaluated using a dataset of experimentally validated IL-2-inducing peptides. Evaluation using independent IL-2 prediction models showed that Autoregressive Transformer combined with Hill Climbing achieved the best overall performance, achieving the mean IL-2 induction confidence score of 0.96 while reducing KL divergence from 2.26 for standalone Hill Climbing to 0.75. A case study on an independent IL-13 inducing peptide dataset showed similar trends, with ART initialized Hill Climbing achieving the mean IL-13 induction score of 0.99 while reducing KL divergence from 1.76 to 0.59. Overall, the framework provides a generalizable approach for balancing functional optimization and distributional realism and can be applied to peptide discovery and data augmentation in imbalanced biological datasets thereby generating high confidence peptides for wet lab validation. HighlightsO_LIProposed a two-phase framework for bioactive peptide generation with potential to address class imbalance in peptide classification tasks. C_LIO_LIPerformed a systematic comparison of distribution-learning and optimization-based approaches for peptide generation. C_LIO_LICombined distribution-learning models for sequence generation with optimization algorithms for improving peptide functional properties. C_LIO_LIDemonstrated the applicability of the proposed framework across multiple bioactive peptide datasets. C_LI

Peptide classification remains challenging in bioinformatics because of limited labeled data, particularly the scarcity of verified negative examples, and the complex relationship between amino acid sequences and biological functions. This study introduces Pep-PU-GAN, a deep learning framework that combines positive-unlabeled (PU) learning, generative adversarial networks (GANs), and graph neural networks (GNNs) for peptide classification. Peptides are represented as sequence-derived residue graphs, with amino acids as nodes and edges connecting adjacent residues, enabling attention-based message passing over local neighborhoods. The architecture includes a generator that produces synthetic peptide embeddings in encoder space and a dual-function discriminator that distinguishes real from synthetic embeddings while performing PU classification. Training uses a custom loss integrating non-negative PU (nnPU) risk estimation with adversarial objectives. A self-training mechanism further incorporates high-confidence synthetic positive embeddings to augment the training set and improve performance. Evaluated on neuropeptide classification using 4,049 positive neuropeptides and 8,558 unlabeled peptides, Pep-PU-GAN outperformed baseline models, achieving an F1 score of 0.93 and an AUROC of 0.98 on an independent held-out benchmark. Pep-PU-GAN provides a promising approach for peptide classification tasks with scarce labeled and abundant unlabeled data, with potential applications in computational biology and drug discovery.

Over the past decade, aptamers have emerged as promising therapeutics, with cancer therapeutics as a major research area. Existing computational methods, however, either predict aptamer-target pairs or design aptamers against a specific target. In this study, we present AntiCapt, a computational method for identifying single-stranded DNA (ssDNA) anticancer aptamers (ACAs) using sequence descriptors and fine-tuned nucleotide language models. The main dataset comprises 1,021 experimentally validated ACAs and an equal number of non-aptamer sequences. Comparative analysis revealed distinct patterns in the nucleotide, dinucleotide, and trinucleotide compositions associated with ACAs. We developed machine learning (ML) models using composition, autocorrelation, binary profiles and structural features. Among these, the best-performing composition-based model achieved an AUC of 0.93 on an independent dataset, while chemical descriptor-and structural feature-based models achieved a maximum AUC of 0.89. ML models using pretrained and fine-tuned NLM embeddings were also developed, with fine-tuned HyenaDNA embeddings achieving the highest performance, with an independent AUC of 0.94. Furthermore, we developed a model for discriminating anticancer aptamers from general aptamers. The best-performing models were integrated into AntiCapt, a web server and a standalone tool for predicting, designing, and genome-scale scanning anticancer aptamers (https://webs.iiitd.edu.in/raghava/anticapt/).

Antimicrobial resistance reduces the effectiveness of conventional antibiotics and has become a major global health threat, highlighting the need for new anti-infective agents. Antimicrobial peptides (AMPs), a diverse class of innate immune effectors with broad-spectrum antimicrobial activity, are promising candidates for combating drug-resistant infections. Identifying AMPs by wet-lab experiments, however, remains costly and time-consuming, creating a strong demand for computational identification methods. Our recently developed method, SAMP, captures region-specific residue distributions based on proportionalized split amino acid composition. However, SAMP might ignore key biochemical information and sequence order information. Here we present SAMP V2, a stacking ensemble learning framework based on biochemical and sequence-order information augmented split amino acid composition (BIA-SAAC), which extends SAMP by integrating pseudo-amino acid composition features with biochemical and sequence-order information into split peptide regions. Specifically, each peptide is divided into N-terminal, middle, and C-terminal regions, and pseudo amino acid composition is calculated within each region. Benchmarking tests on six independent test datasets, SAMP V2 outperformed multiple state-of-the-art models, including AMPpred-MFA and iAMP-Attenpred, in terms of accuracy, MCC, G-measure and F1-score. Given its high and robust performance, SAMP V2 could significantly accelerate the discovery of next-generation antimicrobial therapeutics for addressing the global threat of multidrug-resistant pathogens.

Rapid and accurate identification of viral sequences underpins clinical diagnostics, epidemiological surveillance, and the safety testing of biological products, yet the established alignment-based methods such as BLAST are inherently closed-set, with accuracy tied to how well a query is already represented in the reference database. Recent advances in genomic foundation models (GFMs) have enabled alignment-free approaches to sequence classification. However, their performance relative to traditional methods remains incompletely characterised. We evaluated GFMs, specifically the Nucleotide Transformer (NT) and DNABERT families, for viral sequence classification using a dataset derived from the Reference Viral Database (RVDB). Sequence embeddings were generated with mean, max, and CLS pooling, classified by nearest-neighbour search in an embedding index and compared with BLAST as a reference. Matching BLAST provided a particularly stringent benchmark here, because as a closed-set method it compares each query directly against the exact reference sequences it stores, whereas a foundation model must rely on representations shaped by broad, general-purpose pre-training. On sequences of standard length, the best model, NTv3-650M, reached an accuracy of 0.93, approaching the 0.97 achieved by BLAST. On long genomic sequences, evaluated on a much smaller test set, the difference narrowed further, with NTv3-650M reaching 0.95 against 0.98 for BLAST. Robustness analyses showed that all GFMs were markedly more sensitive to base masking than BLAST. Both methods indexed the same labelled reference sequences, and BLAST compared queries with them directly at the nucleotide level, so approaching it from frozen, general-purpose representations without any parameter update is a meaningful result. NTv3-650M emerged as the most promising candidate for further optimisation and deployment in bioinformatics applications.

1Drug--drug interactions between biotech and small-molecule drugs play a critical role in medication safety and therapeutic efficacy. However, most existing computational DDI prediction methods focus primarily on interactions between small-molecule drugs, leaving biotech-small-molecule interactions comparatively underexplored. In this study, we propose B-SMART-Former, an explainable deep learning framework for predicting interaction types between biotech and small-molecule drugs. The proposed framework integrates ChemBERTa embeddings and Morgan molecular fingerprints for small molecules with ProtBERT embeddings for biotech drugs, eliminating the need for similarity-based features while leveraging complementary molecular representations. These multimodal features are processed by a hybrid architecture that combines Transformer-based self-attention, residual convolutional learning, and a multi-layer perceptron classifier to capture both global contextual dependencies and local discriminative patterns. The model is formulated as a multi-class classification task and evaluated using stratified 10-fold cross-validation. To improve model transparency, Integrated Gradients is employed as a post-hoc explainability method to identify the molecular features that contribute most strongly to each prediction. Experimental results demonstrate that B-SMART-Former achieves a micro-averaged AUROC of 0.9978 and an AUPR of 0.9682 while relying solely on intrinsic molecular representations, remaining competitive with similarity-based approaches. The proposed framework offers an effective and explainable solution for biotech-small-molecule DDI prediction and provides a practical foundation for future computational drug interaction studies.

Interest in bacteriophages as alternative treatments for bacterial infections has grown amid the ongoing crisis of multidrug resistance. However, their narrow host range makes it labor-intensive to identify matching phages for bacterial isolates using conventional experimental methods. Although existing computational models based on biomolecular features have shown promise for phage-host matching, the performance of alternative model settings, including data embeddings, across datasets with different degrees of novelty remains underexplored. This study introduces PhageHost-GNN, a novelty-aware framework using routed models to predict strain-level Klebsiella-phage interactions, evaluated under three clinical scenarios. Using multiple embedding models to encode sequence, structural, and genomic-context features, the framework achieved AUPRCs reaching 0.784, 0.731, and 0.642 in host-unseen, phage-unseen, and both-unseen in silico evaluations, respectively. Benchmarks on experimentally validated datasets showed the clearest advantage over fixed embeddings when both hosts and phages were new to the trained model. In addition to direct inference and de novo training, the framework further implements a fine-tuning pipeline to adapt existing models to local custom datasets while reducing model-training time. Overall, PhageHost-GNN provides a flexible framework for model selection, evaluation, and local adaptation in practical phage-therapy screening.