Yoshihiro Yamanishi

Idiopathic pulmonary fibrosis (IPF) is a chronic and progressive disease with a poor prognosis due to the irreversible loss of lung structure. Unfortunately, distinguishing between this condition and other interstitial pneumonias (IPs) is quite difficult. Hence, we herein present a machine learning method to identify IPF biomarkers by correlating serum extracellular vesicle proteomic data with clinical data for 507 patients with IPs and 39 healthy control subjects. We applied multi‐class sparse correlation analysis to detecting proteins associated with each disease‐specific IP phenotype, and identified surfactant protein B (SFTPB) and BPI fold containing family B member 1 (BPIFB1) as potential IPF biomarker candidates for discriminating IPF from other IPs, with area under the receiver operating characteristic curve scores of 0.679 and 0.678, respectively, outperforming existing biomarkers for IPF. Combinations of SFTPB and BPIFB1 achieved better discriminative performance over the use of SFTPB or BPIFB1 individually. The identified biomarker candidates can be expected to facilitate the noninvasive diagnosis of IPF.

Proteins involved in pathophysiological mechanisms are widely recognized as promising therapeutic targets for drug discovery. The therapeutic effects of drugs are primarily mediated through the inhibition or activation of target proteins. Therefore, beyond identifying the appropriate molecular target, determining the direction of action on the target protein is a critical factor that shapes the efficacy and characteristics of a treatment. In this paper, we present DATTs (Disease‐Associated Therapeutic Targets), which is a new database of therapeutic protein targets with required actions for medical treatment. The information is extracted from public documents such as scientific papers and medical textbooks. The available information includes disease–target protein associations and the actions of drugs on these proteins required for therapeutic effects (e.g., activation, inhibition, or targeting). The DATTs currently includes information on 613 diseases, 1475 therapeutic target proteins, and 7415 disease–protein associations. Additional information such as links to other databases containing data on the molecular functions of target proteins and classification of diseases is also accessible. The DATTs can provide useful clues for investigation of disease mechanisms, novel drug design, and drug repositioning. The DATTs is accessible at https://datts.nibb.ac.jp/ .

Drug discovery inherently involves multiparameter optimization in the molecular design because drug candidate molecules must meet diverse properties such as bioactivity, synthesizability, and pharmacokinetic properties. This optimization has traditionally relied on iterative manual design and experimental testing, which are labor-intensive and time-consuming. There is therefore a strong incentive to develop computational methods that efficiently design drug-like molecules with multiple favorable properties using chemical and biological data on therapeutic targets. This study proposes a novel computational method for multiproperty optimization in the molecular structure design of bioactive molecules using multicontext (i.e., chemically and genetically perturbed) transcriptome data on human cells. We integrate a molecular generative model conditioned on a transcriptome profile observed with the target gene knockdown or overexpression into a reinforcement learning framework, enabling simultaneous optimization of a quantitative estimate of drug-likeness, synthetic accessibility score, and water/octanol partition coefficient, while accounting for system-level biological effects on a therapeutic target. Using comprehensive benchmarking against established baselines and rigorous validation across multiple metrics, we demonstrate that the proposed method consistently yields molecules with more favorable drug-like characteristics than existing methods. This proposed method can help achieve more efficient identification of novel drug candidates.

The generation of molecules with desirable chemical properties presents a critical challenge in fields such as chemical synthesis and drug discovery. Recent advancements in artificial intelligence (AI) and deep learning have significantly contributed to data-driven molecular generation. However, challenges persist due to the inherent sensitivity of simplified molecular input line entry system (SMILES) representations and the difficulty in applying generative adversarial networks (GANs) to discrete data. This study introduces reinforcement learning-driven molecular generative adversarial network (RL-MolGAN), a novel transformer-based discrete GAN framework designed to address these challenges. Unlike traditional transformer architectures, RL-MolGAN uses a first-decoder-then-encoder structure, facilitating the generation of drug-like molecules from both de novo and scaffold-based designs. In addition, RL-MolGAN integrates reinforcement learning (RL) and Monte Carlo tree search (MCTS) techniques to increase the stability of GAN training and optimize the chemical properties of the generated molecules. To further improve the model’s performance, RL-MolWGAN, an extension of RL-MolGAN, incorporates the Wasserstein distance and minibatch discrimination, which together enhance the stability of the GAN. Experimental results from two widely used molecular datasets, namely, QM9 and ZINC, validate the effectiveness of our models in generating high-quality molecular structures with diverse and desirable chemical properties. Our source code is publicly available at: https://github.com/tang777777/MIR .

Direct reprogramming (DR) involves inducing cell differentiation by converting somatic cells directly into target cells without bypassing induced pluripotent stem cells. Small molecule-induced DR reduces tumorigenic risk relative to transcription factor-triggered DR. However, identifying small molecules for DR through experimental exploration alone remains challenging. This study presents a computational method, SuperDIRECTEUR, to predict small molecules for DR using single-cell temporal transcriptome data. The cellular conversion processes in DR were simulated by calculating the RNA velocity in each cell. These processes were then classified into multiple stages (i.e., primal, immature, and mature stages), and a variant of simulated annealing was employed to search for the combination of small molecules that mimic stage-specific gene expression patterns during DR. We applied SuperDIRECTEUR to identifying candidate small molecules for DR converting mouse embryonic fibroblasts to induced neurons, and were able to reproduce experimentally verified and functionally related molecules inducing the corresponding conversions in a temporal stage-specific manner. The target proteins of the predicted small molecules in the early and later stages were distinctively involved in biologically relevant pathways toward neuronal differentiation and maturation. The proposed method has potential for practical applications in regenerative medicine.

Motivation: The detailed molecular mechanisms and disease risk factors for heart failure, especially in the Japanese population, remain to be identified. In this study, we developed a trans-omics approach integrating multi-omics data to explore potential disease risk factors on a genome-wide scale. We functionally annotated the single-nucleotide polymorphisms (SNPs) investigated in a Japanese heart failure genome-wide association study using the epigenome data of cis-regulatory elements and regulome data of the transcription factor-binding regions identified in vascular endothelial cells. Results: rs3176334, located in the promoter region of CDKN1A, and rs12437763, located in an enhancer, were identified as candidate heart failure-associated SNPs in the Japanese population. Furthermore, the downstream genes regulated by the enhancer containing rs12437763 were predicted to be multiple C2 and trans-membrane domain containing two (MCTP2) and nuclear receptor subfamily two group F member two (NR2F2), both of which are known causative genes for congenital heart disease. Availability and implementation: These candidate variants are potential risk factors for heart failure in the Japanese population.

Deep generative models, such as generative adversarial networks (GANs), have been employed for de~novo molecular generation in drug discovery. Most prior studies have utilized reinforcement learning (RL) algorithms, particularly Monte Carlo tree search (MCTS), to handle the discrete nature of molecular representations in GANs. However, due to the inherent instability in training GANs and RL models, along with the high computational cost associated with MCTS sampling, MCTS RL-based GANs struggle to scale to large chemical databases. To tackle these challenges, this study introduces a novel GAN based on actor-critic RL with instant and global rewards, called InstGAN, to generate molecules at the token-level with multi-property optimization. Furthermore, maximized information entropy is leveraged to alleviate the mode collapse. The experimental results demonstrate that InstGAN outperforms other baselines, achieves comparable performance to state-of-the-art models, and efficiently generates molecules with multi-property optimization. The code is available at: https://github.com/tang777777/InstGAN.

Food absorption is dependent on the activities of internal microorganisms. When exploring food functionality, considering the food compounds and their metabolites produced by microbial metabolism is crucial. In this study, we developed a machine learning method to predict food functionalities using microorganism and metabolic data. The prediction was performed on the chemical properties of 70,478 constituent compounds and 24,255 metabolites and their interactions with target proteins in disease-related pathways. This identified potential functional associations between 941 foods and 83 diseases, providing insights into the mechanisms involved, particularly those related to microorganisms. The effects of microorganism-mediated foods on diseases can be categorized based on food type. For example, the microorganisms associated with diseases within each food category indicated the potential involvement of the Bacteroidetes phylum in dyslipidemia and the Firmicutes phylum in Parkinson's disease. This method could aid in identifying disease-associated microorganisms, identifying prebiotic foods, and highlighting the potential of food interventions in the preventive medicine field. Furthermore, this approach is expected to be useful for elucidating novel mechanisms of food-disease interactions mediated by microbial metabolites.