Protein multi-level structure feature-integrated deep learning method for mutational effect prediction.

Through iterative rounds of mutation and selection, proteins can be engineered to enhance their desired biological functions. Nevertheless, identifying optimal mutation sites for directed evolution remains challenging due to the vastness of the protein sequence landscape and the epistatic mutational effects across residues. To address this challenge, we introduce MLSmut, a deep learning-based approach that leverages multi-level structural features of proteins. MLSmut extracts salient information from protein co-evolution, sequence semantics, and geometric features to predict the mutational effect. Extensive benchmark evaluations on 10 single-site and two multi-site deep mutation scanning datasets demonstrate that MLSmut surpasses existing methods in predicting mutational outcomes. To overcome the limited training data availability, we employ a two-stage training strategy: initial coarse-tuning on a large corpus of unlabeled protein data followed by fine-tuning on a curated dataset of 40-100 experimental measurements. This approach enables our model to achieve satisfactory performance on downstream protein prediction tasks. Importantly, our model holds the potential to predict the mutational effects of any protein sequence. Collectively, these findings suggest that our approach can substantially reduce the reliance on laborious wet lab experiments and deepen our understanding of the intricate relationships between mutations and protein function.

Inferring the distributions of fitness effects and proportions of strongly deleterious mutations.

The distribution of fitness effects is a key property in evolutionary genetics as it has implications for several evolutionary phenomena including the evolution of sex and mating systems, the rate of adaptive evolution, and the prevalence of deleterious mutations. Despite the distribution of fitness effects being extensively studied, the effects of strongly deleterious mutations are difficult to infer since such mutations are unlikely to be present in a sample of haplotypes, so genetic data may contain very little information about them. Recent work has attempted to correct for this issue by expanding the classic gamma-distributed model to explicitly account for strongly deleterious mutations. Here, we use simulations to investigate one such method, adding a parameter (plth) to capture the proportion of strongly deleterious mutations. We show that plth can improve the model fit when applied to individual species but underestimates the true proportion of strongly deleterious mutations. The parameter can also artificially maximize the likelihood when used to jointly infer a distribution of fitness effects from multiple species. As plth and related parameters are used in current inference algorithms, our results are relevant with respect to avoiding model artifacts and improving future tools for inferring the distribution of fitness effects.

Towards Quantum-Chemical Level Calculations of SARS-CoV-2 Spike Protein Variants of Concern by First Principles Density Functional Theory.

The spike protein (S-protein) is a crucial part of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), with its many domains responsible for binding, fusion, and host cell entry. In this review we use the density functional theory (DFT) calculations to analyze the atomic-scale interactions and investigate the consequences of mutations in S-protein domains. We specifically describe the key amino acids and functions of each domain, which are essential for structural stability as well as recognition and fusion processes with the host cell; in addition, we speculate on how mutations affect these properties. Such unprecedented large-scale ab initio calculations, with up to 5000 atoms in the system, are based on the novel concept of amino acid-amino acid-bond pair unit (AABPU) that allows for an alternative description of proteins, providing valuable information on partial charge, interatomic bonding and hydrogen bond (HB) formation. In general, our results show that the S-protein mutations for different variants foster an increased positive partial charge, alter the interatomic interactions, and disrupt the HB networks. We conclude by outlining a roadmap for future computational research of biomolecular virus-related systems.

Using machine learning to predict the effects and consequences of mutations in proteins.

Machine and deep learning approaches can leverage the increasingly available massive datasets of protein sequences, structures, and mutational effects to predict variants with improved fitness. Many different approaches are being developed, but systematic benchmarking studies indicate that even though the specifics of the machine learning algorithms matter, the more important constraint comes from the data availability and quality utilized during training. In cases where little experimental data are available, unsupervised and self-supervised pre-training with generic protein datasets can still perform well after subsequent refinement via hybrid or transfer learning approaches. Overall, recent progress in this field has been staggering, and machine learning approaches will likely play a major role in future breakthroughs in protein biochemistry and engineering.

Pinholin S21 mutations induce structural topology and conformational changes.

The bacteriophage infection cycle is terminated at a predefined time to release the progeny virions via a robust lytic system composed of holin, endolysin, and spanin proteins. Holin is the timekeeper of this process. Pinholin S21 is a prototype holin of phage Φ21, which determines the timing of host cell lysis through the coordinated efforts of pinholin and antipinholin. However, mutations in pinholin and antipinholin play a significant role in modulating the timing of lysis depending on adverse or favorable growth conditions. Earlier studies have shown that single point mutations of pinholin S21 alter the cell lysis timing, a proxy for pinholin function as lysis is also dependent on other lytic proteins. In this study, continuous wave electron paramagnetic resonance (CW-EPR) power saturation and double electron-electron resonance (DEER) spectroscopic techniques were used to directly probe the effects of mutations on the structure and conformational changes of pinholin S21 that correlate with pinholin function. DEER and CW-EPR power saturation data clearly demonstrate that increased hydrophilicity induced by residue mutations accelerate the externalization of antipinholin transmembrane domain 1 (TMD1), while increased hydrophobicity prevents the externalization of TMD1. This altered hydrophobicity is potentially accelerating or delaying the activation of pinholin S21. It was also found that mutations can influence intra- or intermolecular interactions in this system, which contribute to the activation of pinholin and modulate the cell lysis timing. This could be a novel approach to analyze the mutational effects on other holin systems, as well as any other membrane protein in which mutation directly leads to structural and conformational changes.

Fitness effects of mutation in natural populations of Arabidopsis thaliana reveal a complex influence of local adaptation.

Little is empirically known about the contribution of mutations to fitness in natural environments. However, Fisher's Geometric Model (FGM) provides a conceptual foundation to consider the influence of the environment on mutational effects. To quantify mutational properties in the field, we established eight sets of MA lines (7-10 generations) derived from eight founders collected from natural populations of Arabidopsis thaliana from French and Swedish sites, representing the range margins of the species in Europe. We reciprocally planted the MA lines and their founders at French and Swedish sites, allowing us to test predictions of FGM under naturally occurring environmental conditions. The performance of the MA lines relative to each other and to their respective founders confirmed some and contradicted other predictions of the FGM: the contribution of mutation to fitness variance increased when the genotype was in an environment where its fitness was low, that is, in the away environment, but mutations were more likely to be beneficial when the genotype was in its home environment. Consequently, environmental context plays a large role in the contribution of mutations to the evolutionary process and local adaptation does not guarantee that a genotype is at or close to its optimum.

Spontaneous mutation accumulation in multiple strains of the green alga, Chlamydomonas reinhardtii.

Estimates of mutational parameters, such as the average fitness effect of a new mutation and the rate at which new genetic variation for fitness is created by mutation, are important for the understanding of many biological processes. However, the causes of interspecific variation in mutational parameters and the extent to which they vary within species remain largely unknown. We maintained multiple strains of the unicellular eukaryote Chlamydomonas reinhardtii, for approximately 1000 generations under relaxed selection by transferring a single cell every ~10 generations. Mean fitness of the lines tended to decline with generations of mutation accumulation whereas mutational variance increased. We did not find any evidence for differences among strains in any of the mutational parameters estimated. The overall change in mean fitness per cell division and rate of input of mutational variance per cell division were more similar to values observed in multicellular organisms than to those in other single-celled microbes. However, after taking into account differences in genome size among species, estimates from multicellular organisms and microbes, including our new estimates from C. reinhardtii, become substantially more similar. Thus, we suggest that variation in genome size is an important determinant of interspecific variation in mutational parameters.

Computer-Aided Protein Directed Evolution: a Review of Web Servers, Databases and other Computational Tools for Protein Engineering.

The combination of computational and directed evolution methods has proven a winning strategy for protein engineering. We refer to this approach as computer-aided protein directed evolution (CAPDE) and the review summarizes the recent developments in this rapidly growing field. We will restrict ourselves to overview the availability, usability and limitations of web servers, databases and other computational tools proposed in the last five years. The goal of this review is to provide concise information about currently available computational resources to assist the design of directed evolution based protein engineering experiment.

PERSPECTIVE: SPONTANEOUS DELETERIOUS MUTATION.

Mildly deleterious mutation has been invoked as a leading explanation for a diverse array of observations in evolutionary genetics and molecular evolution and is thought to be a significant risk of extinction for small populations. However, much of the empirical evidence for the deleterious-mutation process derives from studies of Drosophila melanogaster, some of which have been called into question. We review a broad array of data that collectively support the hypothesis that deleterious mutations arise in flies at rate of about one per individual per generation, with the average mutation decreasing fitness by about only 2% in the heterozygous state. Empirical evidence from microbes, plants, and several other animal species provide further support for the idea that most mutations have only mildly deleterious effects on fitness, and several other species appear to have genomic mutation rates that are of the order of magnitude observed in Drosophila. However, there is mounting evidence that some organisms have genomic deleterious mutation rates that are substantially lower than one per individual per generation. These lower rates may be at least partially reconciled with the Drosophila data by taking into consideration the number of germline cell divisions per generation. To fully resolve the existing controversy over the properties of spontaneous mutations, a number of issues need to be clarified. These include the form of the distribution of mutational effects and the extent to which this is modified by the environmental and genetic background and the contribution of basic biological features such as generation length and genome size to interspecific differences in the genomic mutation rate. Once such information is available, it should be possible to make a refined statement about the long-term impact of mutation on the genetic integrity of human populations subject to relaxed selection resulting from modern medical procedures.