After being on hiatus for a while, we are back with giving you (sometimes) weekly updates!

Structure-informed microbial population genetics elucidate selective pressures that shape protein evolution

This paper showcases how novel scientific directions can be explored thanks to the revolutionary advances in the protein structure prediction field. The authors describe how evolutionary pressures governing the genetic diversity in a marine microbial population can be discerned by analyzing its genetic variation in the context of their respective protein structures models. The paper culminates in the demonstration of how nitrogen availability is responsible for the rate of nonideal polymorphisms at a critical enzymatic site.

Early Selection of the Amino Acid Alphabet Was Adaptively Shaped by Biophysical Constraints of Foldability

The authors address the questions raised by the fact that our well-loved protein alphabet of 20 amino acids was not always so large. In fact, only 10 amino acids are considered “early” (ADQGILPSTV) and found in prebiotic environments. And, in the authors words, “While some of the choices for the early amino acids are quite expected, others have been repeatedly considered thought-provoking”. So, they create libraries of 25-mer peptides for different subsets of amino acids, both early proteogenic and non- and look for differences in solubility and foldability. The early subsets are more soluble than the modern set, supporting the view that modern proteins are more reliant on chaperones and/or translation to fold. However, our modern set has higher secondary structure propensity, meaning more interesting folds. In general, the negatively charged AAs (DQ) tend to be closer to the surface in short peptides and could end up forming ion-pair salt bridges with positive AAs that stabilize unfolded conformations - early proteins seem to have avoided this by not having positive AAs at all, while modern ones used R and K once they became available, which have longer side-chains away from the backbone so that they interact mainly with water. One cool hypothesis: to make a xeno alphabet, you could use positive AAs with short side chains and negative ones with longer side chains and hopefully still get the same compromise between solubility and foldability that we see now.

Science, the hard way

In this opinionated blog post, Joe Greener argues that machine learning papers in structural biology need to do more rigorous validation that better suit the use cases these methods are developed for. He dives in to some specific problems involved in machine learning in forcefields, protein-ligand binding, protein structure prediction and protein design. He concludes with a number of solutions to these problems, and how to do science ‘the hard way’.

Some more