MOTIVATION: Drug-target interaction (DTI) prediction is a crucial step in modern drug discovery. Accurate and efficient predictions can substantially reduce costs and development time. Applications of deep learning methods for this purpose have been extensively studied in recent years, yielding instrumental contributions to this field. However, existing methods face issues pertaining to efficient learning of drug and target feature representations, which is detrimental to generalizability and performance in cold-start scenarios. Most approaches extract representations from SMILES strings for drugs and FASTA sequences for target proteins, which encode limited 3D structural information. Additionally, many models lack explainability, being black boxes that provide little physical insight into the underlying mechanisms behind such interactions. RESULTS: We propose 3DICE, a novel framework leveraging co-attention-based fusion and massively pre-trained 3D structural encoders for both drugs and proteins. Uni-Mol and ESM-IF1 are employed to generate high-fidelity, 3D structure-aware embeddings which enable richer geometric and chemical understanding. Cross-modal fusion modules further augment representations to model intermolecular binding relationships. Importantly, this mechanism also provides intrinsic interpretability, highlighting and enabling qualitative analysis of most influential atoms or residues. Experiments conducted on two canonical benchmark datasets display the competitiveness of our model in real-world scenarios. 3DICE outperformed state-of-the-art models across multiple metrics on the DrugBank and KIBA datasets. Additional experiments provide a more rigorous analysis of interpretability than is typically reported in prior DTI studies, and we find that attention consistently highlights decision-critical regions which is not intrinsically class-specific. AVAILABILITY: Our model and dataset are freely available at: https://github.com/austinatose/3DICE.

MOTIVATION: Foundation models such as scGPT have demonstrated strong potential for single-cell multi-omics integration; however, their downstream performance is highly sensitive to hyperparameter selection. Manual fine-tuning remains computationally expensive, dataset-dependent, and often irreproducible. Despite the increasing adoption of foundation models in single-cell analysis, systematic strategies for robust hyperparameter optimization remain underexplored. RESULTS: We developed a Bayesian optimization framework based on Tree-structured Parzen Estimators (TPE) for automated fine-tuning of scGPT and evaluated its performance on two benchmark bone marrow mononuclear cell (BMMC) multi-omics datasets, including CITE-seq and GSE194122 datasets. Across datasets, Bayesian optimization consistently improved biological conservation and batch integration metrics compared with default scGPT configurations. On the original BMMC benchmark, optimization improved AvgBIO from 0.59 to 0.67 and PCR from 0.33 to 0.52. On the GSE194122 dataset, the default configuration exhibited unstable convergence and weak biological preservation (AvgBIO = 0.19; ARI = 0.007), whereas Bayesian optimization substantially improved integration performance (AvgBIO = 0.60; ARI = 0.63) while reducing validation loss from 137 to 47.1. These findings demonstrate substantial dataset-specific sensitivity of scGPT fine-tuning and highlight the importance of automated optimization for stable deployment across heterogeneous multi-omics datasets. Our study demonstrates that Bayesian optimization provides an effective and reproducible strategy for stabilizing scGPT fine-tuning across diverse single-cell multi-omics datasets. Rather than introducing a new integration architecture, this work emphasizes the importance of systematic optimization for improving robustness and reproducibility of foundation-model applications in computational biology. AVAILABILITY AND IMPLEMENTATION: Our model and dataset are freely available at: https://github.com/daren642/scGPT_multiomic_tuning.

Anticancer peptides (ACPs) are the types of peptides that have been demonstrated to have anticancer activities. Using ACPs to prevent cancer could be a viable alternative to conventional cancer treatments because they are safer and display higher selectivity. Due to ACP identification being highly lab-limited, expensive and lengthy, a computational method is proposed to predict ACPs from sequence information in this study. The process includes the input of the peptide sequences, feature extraction in terms of ordinal encoding with positional information and handcrafted features, and finally feature selection. The whole model comprises of two modules, including deep learning and machine learning algorithms. The deep learning module contained two channels: bidirectional long short-term memory (BiLSTM) and convolutional neural network (CNN). Light Gradient Boosting Machine (LightGBM) was used in the machine learning module. Finally, this study voted the three models' classification results for the three paths resulting in the model ensemble layer. This study provides insights into ACP prediction utilizing a novel method and presented a promising performance. It used a benchmark dataset for further exploration and improvement compared with previous studies. Our final model has an accuracy of 0.7895, sensitivity of 0.8153 and specificity of 0.7676, and it was increased by at least 2% compared with the state-of-the-art studies in all metrics. Hence, this paper presents a novel method that can potentially predict ACPs more effectively and efficiently. The work and source codes are made available to the community of researchers and developers at https://github.com/khanhlee/acp-ope/.

OBJECTIVE: To assess large language models on their ability to accurately infer cancer disease response from free-text radiology reports. MATERIALS AND METHODS: We assembled 10 602 computed tomography reports from cancer patients seen at a single institution. All reports were classified into: no evidence of disease, partial response, stable disease, or progressive disease. We applied transformer models, a bidirectional long short-term memory model, a convolutional neural network model, and conventional machine learning methods to this task. Data augmentation using sentence permutation with consistency loss as well as prompt-based fine-tuning were used on the best-performing models. Models were validated on a hold-out test set and an external validation set based on Response Evaluation Criteria in Solid Tumors (RECIST) classifications. RESULTS: The best-performing model was the GatorTron transformer which achieved an accuracy of 0.8916 on the test set and 0.8919 on the RECIST validation set. Data augmentation further improved the accuracy to 0.8976. Prompt-based fine-tuning did not further improve accuracy but was able to reduce the number of training reports to 500 while still achieving good performance. DISCUSSION: These models could be used by researchers to derive progression-free survival in large datasets. It may also serve as a decision support tool by providing clinicians an automated second opinion of disease response. CONCLUSIONS: Large clinical language models demonstrate potential to infer cancer disease response from radiology reports at scale. Data augmentation techniques are useful to further improve performance. Prompt-based fine-tuning can significantly reduce the size of the training dataset.

With the mindset of constant improvement in efficiency and safety in the workspace and training in Singapore, there is a need to explore varying technologies and their capabilities to fulfil this need. The ability of Virtual Reality (VR) and Augmented Reality (AR) to create an immersive experience of tying the virtual and physical environments coupled with information filtering capabilities brings a possibility of introducing this technology into the training process and workspace. This paper surveys current research trends, findings and limitation of VR and AR in its effect on human performance, specifically in Singapore, and our experience in the National University of Singapore (NUS).

Diabetic retinopathy (DR) is a common retinal disease that leads to blindness. For diagnosis purposes, DR image grading aims to provide automatic DR grade classification, which is not addressed in conventional research methods of binary DR image classification. Small objects in the eye images, like lesions and microaneurysms, are essential to DR grading in medical imaging, but they could easily be influenced by other objects. To address these challenges, we propose a new deep learning architecture, called BiRA-Net, which combines the attention model for feature extraction and bilinear model for fine-grained classification. Furthermore, in considering the distance between different grades of different DR categories, we propose a new loss function, called grading loss, which leads to improved training convergence of the proposed approach. Experimental results are provided to demonstrate the superior performance of the proposed approach.