TrioFold

Description

Here, we propose TrioFold to achieve enhanced generalizability of RSS prediction by integrating base-pairing clues learned from both thermodynamic- and DL-based methods by ensemble learning and convolutional block attention mechanism. TrioFold achieves higher accuracy in intra-family predictions and enhanced generalizability in inter-families and cross-RNA-types predictions. Importantly, TrioFold uses only ~2800 parameters to achieve superior performance over SOTA DL methods requiring millions of parameters. This study demonstrated new opportunities to enhance generalizability for RSS predictions by efficiently ensemble learning of base-pairing clues learned from both thermodynamic- and DL-based algorithms.

Categories

Licence