Recent Trends in Machine Learning-Based Computational Biology and Digital Medicine

Recently, machine learning has rapidly become a major research topic in the range of applications involving computational biology and medical diagnosis. With the improvement and development of medical research databases and biological databases, researchers have explored more and more new questions and corresponding new data. Traditional calculation methods are becoming increasingly unsuitable for these new data. Therefore, there is an increasing need for more new algorithms and models to adapt to updated data. This thematic issue entitled "Recent Trends in Machine Learning-Based Computational Biology and Digital Medicine" aims to explore the scientific and technological frontiers that address key issues and challenges in computational biology and computational medicine. This thematic issue has attracted 12 article submissions. These papers covered the methods, implementation, results, and evaluation of novel approaches and technologies in the field of machine learning-based computational biology and digital medicine. The topics of these papers encompassed e-health, artificial intelligence, biomedical engineering, mathematical optimization, medical image analysis, biomedical discovery, and computational biology. This mini-thematic issue ultimately accepted 3 papers, out of which one review article surveyed recent deep-learning methods for protein prediction, and two research articles presented new models and algorithms for specific biomedical application problems. <br> Only one review article has been included in this mini-thematic issue. Deep learning, a powerful technology for large-scale biological and biomedical data processing, has great application potential in the field of bioinformatics. Luo and Cai [1] have reviewed the recent prediction approaches for proteins. Firstly, they have introduced the background of deep learning. Then they have briefly surveyed the applications of deep learning in protein prediction, including protein sequence analysis, protein structure prediction, protein function prediction, and prediction of protein-protein interactions. Afterward, they have pointed out the challenges, limitations, and risks in deep learning-based protein prediction. In the end, they have discussed the future directions of deep learning-based protein computation. <br> In addition, two research articles have been included in this mini-thematic issue. Deep learning-based analysis of multi-site functional magnetic resonance imaging (fMRI) data is significantly helpful for the treatment of autism spectrum disorder (ASD). The current methods have been found to mainly rely on graph convolutional neural networks (GCNs), but the deep learning-based method cannot describe neighbor information at an adjustable scale. Jiao et al. [2] have, thus, proposed a method to integrate a population-based graph framework and the homogeneous graph wavelet neural network (H-GWNN) by learning the representation of graph classification in an end-to-end manner. They have fused image data and phenotype data based on position and scale effects to construct a graph, and found the optimal scale by adjusting the scale in the graph wavelet neural network (GWNN) to extract the most discriminative features on the graph. This method has been validated on the autism dataset ABIDE, surpassing other methods based on the GCN framework. The methods based on traditional machine learning and deep learning have achieved good results in the diagnosis of COVID-19. However, the effect of the large language model (LLM) in the diagnosis of COVID-19 severity and clinical outcomes still needs to be explored. Zhu et al. [3] have proposed an LLM approach with missing value adaptation for COVID-19 disease severity and clinical outcomes diagnosis. They have set special prompts in the LLM to explicitly notify the LLM when missing values are encountered, without the need to interpolate the missing values. They have also adopted a multi-objective learning strategy to predict disease severity first and then clinical outcomes. In the fine-tuning process of LLM, these two objectives have been mutually promoted. The experimental results have demonstrated this method to exhibit superior performance in predicting disease severity and clinical outcomes. <br> The collected articles have provided an insight into some issues inherent in computational biology and digital medicine, and provided some inspiration for researchers' future work in this field.