Interpretable Pairwise Distillations for Generative Protein Sequence Models

Many different types of generative models for protein sequences have been proposed in literature. Their uses include the prediction of mutational effects, protein design and the prediction of structural properties. Neural network (NN) architectures have shown great performances, commonly attributed to the capacity to extract non-trivial higher-order interactions from the data. In this work, we analyze three different NN models and assess how close they are to simple pairwise distributions, which have been used in the past for similar problems. We present an approach for extracting pairwise models from more complex ones using an energy-based modeling framework. We show that for the tested models the extracted pairwise models can replicate the energies of the original models and are also close in performance in tasks like mutational effect prediction.

The fitness landscapes of translation

Motivated by recent experiments on an antibiotic resistance gene, we investigate genetic interactions between synonymous mutations in the framework of exclusion models of translation. We show that the range of possible interactions is markedly different depending on whether translation efficiency is assumed to be proportional to ribosome current or ribosome speed. In the first case every mutational effect has a definite sign that is independent of genetic background, whereas in the second case the effect-sign can vary depending on the presence of other mutations. The latter result is demonstrated using configurations of multiple translational bottlenecks induced by slow codons.

Thr90Ser Mutation in Antithrombin is Associated with Recurrent Thrombosis in a Heterozygous Carrier

Antithrombin (AT) is a serine protease inhibitor that regulates the activity of coagulation proteases of both intrinsic and extrinsic pathways. We identified an AT-deficient patient with a heterozygous Thr90Ser (T90S) mutation who experiences recurrent venous thrombosis. To understand the molecular basis of the clotting defect, we expressed AT-T90S in mammalian cells, purified it to homogeneity, and characterized its properties in established kinetics, binding, and coagulation assays. The possible effect of mutation on the AT structure was also evaluated by molecular modeling. Results demonstrate the inhibitory activity of AT-T90S toward thrombin and factor Xa has been impaired three- to fivefold in both the absence and presence of heparin. The affinity of heparin for AT-T90S has been decreased by four- to fivefold. Kinetic analysis revealed the stoichiometry of AT-T90S inhibition of both thrombin and factor Xa has been elevated by three- to fourfold in both the absence and presence of heparin, suggesting that the reactivity of coagulation proteases with AT-T90S has been elevated in the substrate pathway. The anticoagulant activity of AT-T90S has been significantly impaired as analyzed in the AT-deficient plasma supplemented with AT-T90S. The anti-inflammatory effect of AT-T90S was also decreased. Structural analysis predicts the shorter side-chain of Ser in AT-T90S has a destabilizing effect on the structure of AT and/or the AT-protease complex, possibly increasing the size of an internal cavity and altering a hydrogen-bonding network that modulates conformations of the allosterically linked heparin-binding site and reactive center loop of the serpin. This mutational effect increases the reactivity of AT-T90S with coagulation proteases in the substrate pathway.

Disentangling the contribution of each descriptive characteristic of every single mutation to its functional effects

Mutational effects predictions continue to improve in accuracy as advanced artificial intelligence (AI) algorithms are trained on exhaustive experimental data. The next natural questions to ask are if it is now possible to gain insights into which attribute of the mutation contributes how much to the mutational effects, and if one can develop universal rules for mapping the descriptors to mutational effects. In this work, we mainly address the former aspect using a framework of interpretable AI. Relations between the physico-chemical descriptors and their contributions to the mutational effects are extracted by analyzing the data on 29,832 variants from 8 systematic deep-mutational scan studies. It is found that the intuitive dependences of fitness and solubility on the distance of the amino acid from active site could be extracted and quantified. The dependence of the mutational effect contributions on the number of contacts an amino acid has or the BLOSUM score descriptor of the change showed universal trends. Our attempts in the present work to explain the quantitative differences in the dependence on conservation and SASA across proteins were not successful. The work nevertheless brings transparency into the predictions, development of rules, and will hopefully lead to uncovering the universalities among these rules.

Silent crickets reveal the genomic footprint of recent adaptive trait loss

Secondary trait loss is widespread and has profound consequences, from generating diversity to driving adaptation. Sexual trait loss is particularly common. Its genomic impact is challenging to reconstruct because most reversals occurred in the distant evolutionary past and must be inferred indirectly, and questions remain about the extent of disruption caused by pleiotropy, altered gene expression and loss of homeostasis. We tested the genomic signature of recent sexual signal loss in Hawaiian field crickets, Teleogryllus oceanicus. Song loss is controlled by a sex-linked Mendelian locus, flatwing, which feminises male wings by erasing sound-producing veins. This variant spread rapidly under pressure from an eavesdropping parasitoid fly. We sequenced, assembled and annotated the T. oceanicus genome, produced a high-density linkage map, and localised flatwing on the X chromosome. We characterised pleiotropic effects of flatwing, including changes in embryonic gene expression and alteration of another sexual signal, chemical pheromones. Song loss is associated with pleiotropy, hitchhiking and genome-wide regulatory disruption which feminises flatwing male pheromones. The footprint of recent adaptive trait loss illustrates R. A. Fisher's influential prediction that variants with large mutational effect sizes can invade genomes during the earliest stages of adaptation to extreme pressures, despite having severely disruptive genomic consequences.

SNP analysis implicates role of cytosine methylation in introducing consequential mutations inVibrio choleraegenomes

Epigenetic modifications play a key role in gene regulation and in recognition of self DNA in bacteria. In-spite of their positive role in cell survival, modifications like cytosine methylation incur a mutational cost. Cytosine methylation, specifically 5-methylcytosine, is prone to hydrolytic deamination which leads to C → T and G → A transitions. Here, we first study the abundance of mutagenic cytosine methylation target motifs and show that bacteria likeVibrio choleraemight use motif avoidance as a strategy to minimize the mutational effect of deamination of methylated cytosine. Second by performing SNP analysis on whole genome sequence data fromVibrio choleraepatient isolates we show a) high abundance of cytosine methylation-dependent mutations in the cytosine methylation target motif RCCGGY, b) 95% of these C → T and G → A transitions in the coding region lead to non-synonymous substitutions and c) many of these transitions are associated with membrane proteins and are implicated in virulence. Thus, our SNP analysis ofV. choleraegenomes implicates the role of cytosine methylation in generating genotypic diversity with adaptive potential.