[Paper Review] SentRNA: Improving computational RNA design by incorporating a prior of human design strategies

Jade Shi, Rhiju Das+1 more|arXiv (Cornell University)|Mar 8, 2018

RNA and protein synthesis mechanisms16 references15 citations

TL;DR

SentRNA introduces a deep learning approach that improves computational RNA design by training a fully-connected neural network on human-designed RNA sequences from the EteRNA game. By incorporating human design strategies as a prior, SentRNA achieves state-of-the-art performance on complex RNA targets previously unsolvable by machine algorithms, demonstrating that human intuition can significantly enhance machine learning in RNA design.

ABSTRACT

Solving the RNA inverse folding problem is a critical prerequisite to RNA design, an emerging field in bioengineering with a broad range of applications from reaction catalysis to cancer therapy. Although significant progress has been made in developing machine-based inverse RNA folding algorithms, current approaches still have difficulty designing sequences for large or complex targets. On the other hand, human players of the online RNA design game EteRNA have consistently shown superior performance in this regard, being able to readily design sequences for targets that are challenging for machine algorithms. Here we present a novel approach to the RNA design problem, SentRNA, a design agent consisting of a fully-connected neural network trained end-to-end using human-designed RNA sequences. We show that through this approach, SentRNA can solve complex targets previously unsolvable by any machine-based approach and achieve state-of-the-art performance on two separate challenging test sets. Our results demonstrate that incorporating human design strategies into a design algorithm can significantly boost machine performance and suggests a new paradigm for machine-based RNA design.

Motivation & Objective

• To address the persistent challenge of designing complex or large RNA sequences using machine learning.
• To leverage human-designed RNA sequences from the EteRNA game as a source of effective design strategies.
• To develop a machine learning model that learns from human intuition to improve performance on difficult RNA inverse folding problems.
• To demonstrate that incorporating human design priors can significantly enhance computational RNA design beyond current state-of-the-art methods.

Proposed method

• A fully-connected neural network is trained end-to-end on a dataset of human-designed RNA sequences collected from the EteRNA online game.
• The model learns to map RNA secondary structure targets to stable, functional RNA sequences by learning from human design patterns.
• The training process uses a differentiable loss function that evaluates sequence stability and target structure fidelity.
• The approach treats RNA design as a sequence-to-structure generation problem, with the network learning to generate sequences that fold into desired structures.
• The model is evaluated on two challenging test sets, including previously intractable targets.
• Human design strategies are encoded implicitly through the data distribution, without explicit rule engineering.

Experimental results

Research questions

• RQ1 Can human-designed RNA sequences serve as an effective prior to improve machine learning models in RNA design?
• RQ2 Can a neural network trained on human designs outperform existing machine-based RNA design algorithms on complex or large RNA targets?
• RQ3 To what extent do human design strategies, when encoded in a deep learning model, enhance the stability and accuracy of predicted RNA sequences?
• RQ4 Can a data-driven approach using human intuition surpass purely algorithmic or physics-based RNA design methods?

Key findings

• SentRNA successfully designs sequences for RNA targets that were previously unsolvable by any machine-based approach.
• The model achieves state-of-the-art performance on two challenging test sets, outperforming existing computational RNA design methods.
• By learning from human-designed sequences, SentRNA captures non-trivial design strategies that are difficult to encode explicitly in traditional algorithms.
• The incorporation of human design priors significantly improves the model’s ability to generate stable, functional RNA sequences for complex secondary structures.