Differentiable RNA Secondary Structure Extraction for Deep Learning

2026-09-24

AI Summary

This study addresses the performance bottleneck in existing RNA structure prediction caused by the inconsistency between post-processing extraction algorithms and training objectives. We systematically analyze the alignment between training and extraction methods, comparing dynamic programming, graph matching, and greedy algorithms. To eliminate this discrepancy, we introduce a differentiable SDSM normalization technique that enables the model to directly output probability matrices. Experimental results demonstrate that the SDSM model significantly outperforms baselines across various extraction algorithms. Furthermore, its pre-extraction outputs most closely approximate the ground-truth structures, thereby validating the effectiveness and robustness of the proposed approach.

Abstract

Many deep learning approaches to RNA secondary structure prediction have recently been proposed. They typically output a weight matrix $W$ where $W_{ij}$ is an arbitrary weight for base $i$ pairing with base $j$. Converting this matrix to a predicted secondary structure or base-pairing probability matrix typically involves ad hoc and problematic downstream algorithms. Despite the importance of this conversion step, which we refer to as structure extraction, it has received relatively little attention in the literature. In this work, we analyze how the congruence between training and extraction methods affects prediction performance. To do this, we compare four extraction algorithms: a Nussinov-like dynamic programming method, maximum-weight graph matching and the greedy extraction algorithms used by SPOT-RNA and RiNALMo. These are evaluated on outputs from the pretrained RiNALMo model and three toy models trained in this paper: a differentiable Nussinov-like model, a binary cross-entropy (BCE) baseline, and a model that incorporates a novel symmetric doubly stochastic matrix (SDSM) normalization algorithm during training which allows it to output base-pairing probability matrices directly, without a separate extraction step. This SDSM normalization algorithm is differentiable and can be added inline to any deep learning model during training and evaluation. We find that the performance of each extraction method depends strongly on how the corresponding model was trained. Considering the toy models themselves, the SDSM model showed the strongest overall performance: it outperformed the BCE baseline under all four extraction algorithms and produced pre-extraction outputs closest to the ground truth. These results suggest that SDSM normalization is a tractable alternative to traditional structure extraction.

Problem

• RNA secondary structure prediction
• structure extraction
• deep learning
• base-pairing probability matrix

Innovation

• Differentiable structure extraction
• Symmetric doubly stochastic matrix
• RNA secondary structure prediction
• Deep learning
• SDSM normalization