Jhawar, K. | bioRxiv Science

bioRxiv2025

MkAtt-SDN2GO: Multi-kernel Attentive-SDN2GO Network for Protein Function Prediction in Humans

Accurately annotating the functions of uncharacterised human proteins remains a major bottleneck in biology. We present MkAtt-SDN2GO, a neural architecture that extends SDN2GO by integrating adaptive multi-kernel convolution and attention mechanisms to predict Gene Ontology terms from protein sequences, domains, and protein-protein interaction (PPI) context. The sequence stream employs a learnable multi-kernel convolution layer that combines features from multiple kernel sizes through attention-based gating, enabling adaptive motif detection without relying on a fixed receptive field. A self-attention layer models long-range dependencies, while cross-attention integrates sequence, domain, and PPI representations into a unified prediction space. On a CAFA-style benchmark, MkAtt-SDN2GO improves Molecular Function (MF) Fmax by 14.8% (0.657 vs 0.572) and Recallmax by 18.8% over SDN2GO. Across Homo sapiens, the fused model achieves top Fmax scores in Biological Process (BP) (0.441), MF (0.657), and Cellular Component (CC) (0.522) compared with other methods. Although the domain-only stream performs strongly, the cross-attention fusion enhances robustness and interpretability when individual modalities are weak or missing. Overall, adaptive multi-scale convolution combined with attention thus advances large-scale protein annotation and offers a scalable and potential tool for functional genomics and disease research.

bioinformatics↗

bioRxiv2025

Morphology-guided convolutional Graph Neural Network decodes optically barcoded nanoparticles for one-pot, purification-,amplification-, and enzyme-free femtomolar nucleic-acid diagnostics

Affordable, accurate, and rapid point-of-care diagnostic tests remain elusive due to inherent trade-offs between performance and cost. Conventional nucleic acid tests offer high sensitivity but require complex, expensive steps such as amplification and purification, whereas lateral flow assays are simple and low-cost but lack the necessary sensitivity for many applications. To bridge this gap, we present a miniaturized and simplified chip-based platform that combines three components into a single diagnostic pipeline: we use (i) spectrally distinct silver and gold nanoparticles that form analyte-dependent clusters with unique spectral fingerprints, (ii) a one-pot, enzyme- and purification-free assay on a chip integrated with a high-throughput automated low-cost microscope, and (iii) a morphology-guided convolutional Graph Neural Network that embeds morphology information into convolutional kernels and performs graph-based relational learning across particle-level features. This integration captures spectral, spatial, and morphological quantification at the particle level, rather than relying on bulk spectral shifts, thereby overcoming the limitations of contemporary nanoparticle assays and image-level deep learning approaches. Processing up to 5000 particles per image using only <5 GB GPU memory, Mc-GNN achieves femtomolar sensitivity with 98.2% recall for synthetic DNA and 94.8% for SARS-CoV-2 RNA from whole virus, despite variations in nanoparticle selection and sample complexity. By embedding morphological information into the biosensing pipeline, our diagnostic platform is computationally efficient, smartphone-compatible and is readily extensible to new analytes and multiplexing, offering a scalable solution for a fieldable diagnostic tool.

bioengineering↗