Biology subjects

Umetsu, M.

Publications and source records attributed to Umetsu, M..

6 recordsLinked to original sources

bioRxiv2026

pLM-bVAE: protein language model-enhanced binary representation improves quantum annealing-based protein sequence design

Quantum annealing (QA) has been explored as an approach for protein design in vast combinatorial sequence spaces. Applying QA to protein sequence design requires amino acid sequences to be encoded as binary vectors, and optimized vectors to be decoded back into amino acid sequences. Thus, both the quality of binary encoding and sequence reconstruction accuracy are important for effective QA-based optimization. Binary variational autoencoder (bVAE) has been used to map protein sequences to binary representations in QA-based protein sequence design. However, bVAE was originally developed for natural language modeling, and its suitability for proteins has not been systematically evaluated. Here, we present pLM-bVAE, which incorporates protein representations derived from a pretrained protein language model (pLM) into the original bVAE model. In computational experiments on three public protein variant datasets, pLM-bVAE showed higher sequence reconstruction accuracy and binary encoding performance than bVAE. Moreover, we incorporated pLM-bVAE into a QA-based protein sequence design framework. Across all three datasets, pLM-bVAE generated more high-scoring sequences with greater sequence dissimilarity from the training and validation data than bVAE. Our results demonstrate that pLM-bVAE provides an effective binary representation for improving QA-based protein sequence design. Availability The source code used in this manuscript is available at https://github.com/hueda8/pLM-bVAE.

bioinformatics↗

bioRxiv2026

Proximity directs microbial transglutaminase site selectivity in native antibody modification

Site-specific protein modification allows diverse functionalities to be introduced while minimizing perturbations to the protein structure and activity. Considerable efforts have been made to achieve site-specific modification of native proteins to overcome the heterogeneity resulting from conventional stochastic Lys or Cys modification. We have previously achieved the selective modification of Lys65 in a native immunoglobulin G1 (IgG1) antibody (trastuzumab) using EzMTG-pG(Fab), which is an engineered zymogen of microbial transglutaminase (EzMTG) fused to a Fab-binding protein G [pG(Fab)]. However, this approach cannot be widely applied to different types of IgG antibodies. Here, we designed pG(Fab)-EzMTG by fusing pG(Fab) to the N-terminus of EzMTG. Notably, switching the fusion partners dramatically altered the IgG modification site from Lys65 to Lys225, which is located in the hinge site of native IgG1 antibodies. This Lys225-selective labeling was applicable to different IgG1 antibodies. As a functional application, the cytotoxic drug monomethyl auristatin E (MMAE) was conjugated to Lys225 of trastuzumab, and the resulting antibody-drug conjugate exhibited antigen-specific cytotoxicity. These findings demonstrate that fusion-protein architecture determines site selectivity in proximity-directed enzymatic modification, providing a strategy for the site-specific functionalization of native antibodies.

bioengineering↗

bioRxiv2026

Evaluating Lightweight and Full Fine-Tuning Strategies Against Classical Machine Learning for Protein Function Prediction

Motivation Protein language models (PLMs) have emerged as powerful tools for sequence-based prediction of protein function, yet systematic benchmarks comparing frozen embeddings, fine-tuning strategies like Low-Rank Adaptation (LoRA) and classical machine learning (ML) remain limited. We benchmarked four ML strategies: ML using amino acid descriptors (SL-AAFeat), ML using frozen embeddings from 20 PLMs across various pooling strategies (SL-Embed), full model fine-tuning (FT-Full) and LoRA-based fine-tuning (FT-LoRA). Performance was evaluated on the in-house VHH phage display dataset (VHH) for binding affinity prediction and the TAPE fluorescence dataset (FLS and FLS10) for mutational effect prediction. Results Model performance depended strongly on the dataset and adaptation strategy. Max pooling consistently improved embedding-based models, while amino acid descriptors remained competitive under specific datasets and resource constraints. Fine-tuning generally provided the highest predictive performance, but the advantage is not universal. Hyperparameter optimization significantly enhanced FT-LoRA, enabling it to outperform FT-Full on the VHH dataset with less than 10% model parameter adaptation. In contrast, FT-Full achieved the best performance on FLS and FLS10. Several medium-sized PLMs performed comparably to larger models, highlighting favorable performance-efficiency trade-offs. Overall, this paper presents a thorough review of PLM utilization strategies and practical recommendations for selecting suitable strategies based on dataset characteristics and available computational resources. Availability The source code used in this manuscript is available in a Zenodo repository at https://doi.org/10.5281/zenodo.21466255.

bioinformatics↗

bioRxiv2026

Exploring diverse routes to high-affinity-antibody variable domains through deep-sequencing-informed machine learning

The integration of in vitro selection, deep sequencing, and machine learning (ML) has recently been developed as a powerful strategy for discovering functional antibodies. However, how training data composition and ML search space design influence the identification of high-affinity variants remains unclear. Here, we aimed to optimize ML-integrated directed evolution for functional antibody discovery by selecting training data from deep sequencing analysis. By performing phage display selection using camelid heavy-chain antibodies (VHHs), we demonstrated that early-round data, retaining more binding-negative variants, can be superior for training models to identify high-performance VHHs. We also investigated a lead-independent ML search space design by focusing on conserved residues in final rounds, successfully identifying variants with higher affinities than those from lead-based maturation (KD = 7.9 nM). These findings demonstrate that training data selection and search space design are critical for successful ML-guided antibody engineering and provide diverse pathways for discovering high-affinity VHH variants.

bioengineering↗

bioRxiv2021

Machine-learning-guided library design cycle for directed evolution of enzymes: the effects of training data composition on sequence space exploration

Machine learning (ML) is becoming an attractive tool in mutagenesis-based protein engineering because of its ability to design a variant library containing proteins with a desired function. However, it remains unclear how ML guides directed evolution in sequence space depending on the composition of training data. Here, we present a ML-guided directed evolution study of an enzyme to investigate the effects of a known "highly positive" variant (i.e., variant known to have high enzyme activity) in training data. We performed two separate series of ML-guided directed evolution of Sortase A with and without a known highly positive variant called 5M in training data. In each series, two rounds of ML were conducted: variants predicted by the first round were experimentally evaluated, and used as additional training data for the second-round prediction. The improvements in enzyme activity were comparable between the two series, both achieving enzyme activity 2.2-2.5 times higher than 5M. Intriguingly, the sequences of the improved variants were largely different between the two series, indicating that ML guided the directed evolution to the distinct regions of sequence space depending on the presence/absence of the highly positive variant in the training data. This suggests that the sequence diversity of improved variants can be expanded not only by conventional ML using the whole training data, but also by ML using a subset of the training data even when it lacks highly positive variants. In summary, this study demonstrates the importance of regulating the composition of training data in ML-guided directed evolution.

bioengineering↗

bioRxiv2018

A machine-learning-guided mutagenesis platform for accelerated discovery of novel functional proteins

Molecular evolution based on mutagenesis is widely used in protein engineering. However, optimal proteins are often difficult to obtain due to a large sequence space that requires high costs for screening experiments. Here, we propose a novel approach that combines molecular evolution with machine learning. In this approach, we conduct two rounds of mutagenesis where an initial library of protein variants is used to train a machine-learning model to guide mutagenesis for the second-round library. This enables to prepare a small library suited for screening experiments with high enrichment of functional proteins. We demonstrated a proof-of-concept of our approach by altering the reference green fluorescent protein (GFP) so that its fluorescence is changed to yellow while improving its fluorescence intensity. Using 155 and 78 variants for the initial and the second-round libraries, respectively, we successfully obtained a number of proteins showing yellow fluorescence, 12 of which had better fluorescence performance than the reference yellow fluorescent protein (YFP). These results show the potential of our approach as a powerful platform for accelerated discovery of functional proteins.

bioengineering↗