• Jing Wui Yeoh

Genome-scale metabolic models (GSM) underpin pathway and strain engineering by enabling systematic interventions of cell metabolism for pathway design and host optimization in bioproduction. They support constraint-based analyses such as flux balance analysis, which inform pathway reconstruction, knockout analysis and other engineering strategies to guide experimental design aimed at improving yields of target chemicals.

Despite their central role, practical challenges remain; constructing an efficient and robust implementation workflow is technically demanding, often require specialized expertise, rigorous feasibility checks, and integrated simulation tool chains. Large language models (LLMs) have emerged as powerful assistants for scientific work, offering natural-language interfaces that can explain concepts, parse files, and generate code and documentation, which could lower the barrier to GSM interpretation and analysis workflow setup, accelerating hypothesis generation and improving accessibility for non-experts. Yet, there is limited evidence regarding LLMs’domain knowledge for interpreting GSM and implementing the analysis tasks.

In this work, we present a comprehensive evaluation of LLM capabilities for understanding and analyzing GSMs in metabolic engineering. We have systematically evaluated five main areas: (i) domain knowledge, (ii) metabolic flux prediction, (iii) model reconstruction for pathway, (iv) up- or down-regulation analysis for pathway optimization, and (v) knockout analysis. We benchmarked four prominent LLMs (GPT, Gemini, Claude, and Deepseek-R1) and assessed their outputs using standardized rubric-based scoring metrics, with independent evaluations by all these models. We identified recurrent failure modes and task- and model-specific limitations and articulated best practices for deploying LLMs within GSM workflow and integrating them for knowledge extraction and analysis implementation pipelines in metabolic engineering. This work establishes an evidence-based baseline for LLM-enabled GSM analysis and informs the development of more reliable, accessible, and automation-ready computational workflow for pathway and strain design.

• Yao Ting Huang

Accurate detection of somatic variants is crucial for precision oncology, and long-read sequencing offers unprecedented advantages in resolving complex cancer genomes. However, most long-read somatic callers rely on phasing built for a diploid genome, an assumption violated by various contamination, subclonal heterogeneity, and aneuploidy in tumors. We present LongPhase‑S and LongPhase-TO, two novel methods that jointly reconstructs somatic haplotypes, infers tumor purity from phased haplotypes, and recalibrates somatic variants in a purity‑aware manner using long‑read sequencing. By anchoring each somatic read to a parental germline lineage, LongPhase-S and LongPhase-TO provide a phase-resolved view in which germline and somatic reads are disentangled across the genome. Building on somatic haplotyping, we trained c va tumor-purity predictor that outperformed existing methods across various datasets, particularly at low purity. We showed that LongPhase-S and LongPhase-TO boosted the accuracy of state-of-the-art somatic callers (ClairS and DeepSomatic) by using the estimated purity and somatic haplotypes. Collectively, our findings demonstrated that accurate somatic haplotyping not only improves purity estimation and variant calling but also enhances the resolution of clonal evolution and intratumor heterogeneity.

• Ka-Lok Ng

Quantum Machine Learning (QML) integrates the principles of quantum computing and machine learning, offering promising capabilities for tackling complex computational problems. In this presentation, I will provide an overview of the current state of quantum computing and discuss a study demonstrating quantum advantage in a biomedical application. Building upon the Neural Quantum Embedding (NQE) framework, we observed improved classification accuracy when using Quantum Support Vector Classifiers (QSVC) and Quantum Neural Networks (QNN) compared to their classical counterparts, SVC and NN, for binary classification tasks. To enhance computational efficiency, we integrate Tensor Network (TN) techniques into the QSVC implementation. Our results demonstrate that the NQE+TN+QSVC model not only achieves a higher F1 score, but also reduces training time by more than 20-fold compared to the baseline QSVC. In conclusion, we have developed a robust approach that improves both classification performance and simulation efficiency.

• Somayyeh Koohi

The rapid expansion of genomic data, fueled by advances in sequencing technologies, has outpaced the capabilities of traditional computational methods for sequence analysis. Tasks such as genome comparison, alignment, and similarity scoring are foundational to molecular biology and medicine, yet they remain computationally demanding when applied to large and complex datasets. Conventional alignment-based approaches offer high accuracy but suffer from scalability issues, while alignment-free methods provide speed at the cost of precision and biological interpretability. To overcome these limitations, optical processing has emerged as a promising paradigm for biological data analysis. Leveraging the inherent parallelism and high throughput of optical systems, this approach enables ultra-fast computation with minimal energy consumption and memory requirements. Optical techniques—such as free-space correlation, holography, and Fourier optics—can be harnessed to encode, compare, and visualize genomic sequences in novel ways, facilitating both global and local similarity detection, rearrangement analysis, and evolutionary studies. This talk explores the transformative potential of optical processing in bioinformatics, highlighting its ability to bridge the gap between speed and accuracy in genome analysis. We discuss recent developments in optical architectures for biological data, their application to large-scale sequence comparison, and their implications for future research in genomics, personalized medicine, and evolutionary biology. Optical processing not only accelerates computation but also opens new avenues for biologically meaningful insights, making it a critical frontier in the era of big biological data.

• TBD