Ghassemi, M. | Search bioRxiv

bioRxiv2024

Unifying the Research Landscape of Desiccation Tolerance to Identify Trends, Gaps, and Opportunities

Desiccation tolerance, or the ability to survive extreme dehydration, has evolved recurrently across the tree of life. While our understanding of the mechanisms underlying desiccation tolerance continues to expand, the compartmentalization of findings by study system impedes progress. Here, we analyzed 5,963 papers related to desiccation and examined model systems, research topics, citation networks, and disciplinary siloing over time. Our results show significant siloing, with plant science dominating the field, and relatively isolated clustering of plants, animal, microbial, and fungal literature. Topic modeling identified 46 distinct research topics, highlighting both commonalities and divergences across the knowledge of desiccation tolerance in different systems. We observed a rich diversity of model desiccation tolerant species within the community, contrasting the single species model for most biology research areas. To address citation gaps, we developed a rule-based algorithm to recommend new invitees to a niche conference, DesWorks, enhancing the integration of diverse research areas. The algorithm, which considers co-citation, co-authorship, research topics, and geographic data, successfully identified candidates with novel expertise that was unrepresented in previous conferences. Our findings underscore the importance of interdisciplinary collaboration in advancing desiccation tolerance research and provide a framework for using bibliometric tools to foster scientific integration.

plant biology↗

bioRxiv2020

Self-Supervised Contrastive Learning of Protein Representations By Mutual Information Maximization

Pretrained embedding representations of biological sequences which capture meaningful properties can alleviate many problems associated with supervised learning in biology. We apply the principle of mutual information maximization between local and global information as a self-supervised pretraining signal for protein embeddings. To do so, we divide protein sequences into fixed size fragments, and train an autoregressive model to distinguish between subsequent fragments from the same protein and fragments from random proteins. Our model, CPCProt, achieves comparable performance to state-of-the-art self-supervised models for protein sequence embeddings on various downstream tasks, but reduces the number of parameters down to 2% to 10% of benchmarked models. Further, we explore how downstream assessment protocols affect embedding evaluation, and the effect of contrastive learning hyperparameters on empirical performance. We hope that these results will inform the development of contrastive learning methods in protein biology and other modalities.

bioinformatics↗