Predicting higher-order mutational effects in an RNA enzyme by machine learning of high-throughput experimental data

Abstract

Ribozymes are RNA molecules that catalyze biochemical reactions. Self-cleaving ribozymes are a common naturally occurring class of ribozymes that catalyze site-specific cleavage of their own phosphodiester backbone. In addition to their natural functions, self-cleaving ribozymes have been used to engineer control of gene expression because they can be designed to alter RNA processing and stability. However, the rational design of ribozyme activity remains challenging, and many ribozyme-based systems are engineered or improved by random mutagenesis and selection (in vitro evolution). Improving a ribozyme-based system often requires several mutations to achieve the desired function, but extensive pairwise and higher-order epistasis prevent a simple prediction of the effect of multiple mutations that is needed for rational design. Recently, high-throughput sequencing-based approaches have produced data sets on the effects of numerous mutations in different ribozymes (RNA fitness landscapes). Here we used such high-throughput experimental data from variants of the CPEB3 self-cleaving ribozyme to train a predictive model through machine learning approaches. We trained models using either a random forest or long short-term memory (LSTM) recurrent neural network approach. We found that models trained on a comprehensive set of pairwise mutant data could predict active sequences at higher mutational distances, but the correlation between predicted and experimentally observed self-cleavage activity decreased with increasing mutational distance. Adding sequences with increasingly higher numbers of mutations to the training data improved the correlation at increasing mutational distances. Systematically reducing the size of the training data set suggests that a wide distribution of ribozyme activity may be the key to accurate predictions. Because the model predictions are based only on sequence and activity data, the results demonstrate that this machine learning approach allows readily obtainable experimental data to be used for RNA design efforts even for RNA molecules with unknown structures. The accurate prediction of RNA functions will enable a more comprehensive understanding of RNA fitness landscapes for studying evolution and for guiding RNA-based engineering efforts.

RNA sequence to structure analysis from comprehensive pairwise mutagenesis of multiple self-cleaving ribozymes

“…Instances of positive epistasis are shaded blue, and negative epistasis is shaded red, with higher color intensity indicating a greater magnitude of epistasis. Catalytic residues are indicated by stars along the axes ( A is reproduced from Figure 1B from Beck et al, 2022 ). ( B ) Secondary structure of the CPEB3 ribozyme used in this study.…”

Discovering pathways through ribozyme fitness landscapes using information theoretic quantification of epistasis

“…Machine learning approaches have been applied to the problem of predicting active sequences by extrapolation from mutational data. For example, a random forest model was applied to predict active mutants of a self-cleaving ribozyme ( Breiman 2001 ; Beck et al 2022 ). While often successful in generating predictions, random forest models average over many decision trees and thereby create a difficulty in interpreting the process itself.…”

“…While often successful in generating predictions, random forest models average over many decision trees and thereby create a difficulty in interpreting the process itself. Deep learning models, such as multilayer perceptrons or Long-Short Term Memory networks ( Schmidt and Smolke 2021 ; Beck et al 2022 ; Rotrattanadumrong and Yokobayashi 2022 ), improve the representation of the data and extract features found to be significant to the endpoint being modeled. However, deep models are complex, requiring many parameters, and interpretability remains an unsolved problem ( Kirboga et al 2023 ).…”

Machine Learning-Assisted Engineering of Light, Oxygen, Voltage Photoreceptor Adduct Lifetime

“…Thus, based on the limited data available, the rational selection of combinatorial mutations that tune the dark recovery in a predictable fashion is challenging. Besides, the selection of identified variants for rational recombination often relies on the assumption of simple additive effects on protein fitness . Protein fitness here refers to any protein properties essential for a protein to perform its intended functions effectively.…”

“…Besides, the selection of identified variants for rational recombination often relies on the assumption of simple additive effects on protein fitness. 52 Protein fitness here refers to any protein properties essential for a protein to perform its intended functions effectively. This fitness is a measure of how well a given protein performs a target function and not the typical organismal fitness concept used in evolutionary biology.…”