Predicting water hyacinth expansion and identifying environmental drivers in Lake Tana using machine learning

Predicting capsid protein binding sites in ssRNA viruses relies heavily on computational methods due to labor-intensive experimental techniques. Researchers present a sequence-based framework that combines RNA tertiary modeling, geometric feature extraction, and machine learning. The Qbeta bacteriophage serves as a proof-of-concept system for benchmarking against experimentally identified binding and non-binding RNA fragments.

A neural network trained on geometric descriptors shows high accuracy in distinguishing binding sites, achieving an AUC of 0.88. Applying the model to AlphaFold-predicted RNA structures also maintains predictive performance, though failed predictions suggest additional dynamic factors beyond static conformations influence viral genome packaging.

This study offers insights into RNA-capsid interactions and lays groundwork for applying the approach to other ssRNA viruses.