AutoPeptideML: a study on how to build more trustworthy peptide bioactivity predictors.

MOTIVATION: Automated machine learning (AutoML) solutions can bridge the gap between new computational advances and their real-world applications by enabling experimental scientists to build their own custom models. We examine different steps in the development life-cycle of peptide bioactivity binary predictors and identify key steps where automation cannot only result in a more accessible method, but also more robust and interpretable evaluation leading to more trustworthy models. RESULTS: We present a new automated method for drawing negative peptides that achieves better balance between specificity and generalization than current alternatives. We study the effect of homology-based partitioning for generating the training and testing data subsets and demonstrate that model performance is overestimated when no such homology correction is used, which indicates that prior studies may have overestimated their performance when applied to new peptide sequences. We also conduct a systematic analysis of different protein language models as peptide representation methods and find that they can serve as better descriptors than a naive alternative, but that there is no significant difference across models with different sizes or algorithms. Finally, we demonstrate that an ensemble of optimized traditional machine learning algorithms can compete with more complex neural network models, while being more computationally efficient. We integrate these findings into AutoPeptideML, an easy-to-use AutoML tool to allow researchers without a computational background to build new predictive models for peptide bioactivity in a matter of minutes. AVAILABILITY AND IMPLEMENTATION: Source code, documentation, and data are available at https://github.com/IBM/AutoPeptideML and a dedicated web-server at http://peptide.ucd.ie/AutoPeptideML. A static version of the software to ensure the reproduction of the results is available at https://zenodo.org/records/13363975.

Molecular Dynamics Simulations of Substrate Release from Trypanosoma cruzi UDP-Galactopyranose Mutase.

The enzyme UDP-galactopyranose mutase (UGM) represents a promising drug target for the treatment of infections with Trypanosoma cruzi. We have computed the Potential of Mean Force for the release of UDP-galactopyranose from UGM, using Umbrella Sampling simulations. The simulations revealed the conformational changes that both substrate and enzyme undergo during the process. It was determined that the galactopyranose portion of the substrate is highly mobile and that the opening/closing of the active site occurs in stages. Previously uncharacterized interactions with highly conserved residues were also identified. These findings provide new pieces of information that contribute to the rational design of drugs against T. cruzi.

Steric Control of the Rate-Limiting Step of UDP-Galactopyranose Mutase.

Galactose is an abundant monosaccharide found exclusively in mammals as galactopyranose (Gal p), the six-membered ring form of this sugar. In contrast, galactose appears in many pathogenic microorganisms as the five-membered ring form, galactofuranose (Gal f). Gal f biosynthesis begins with the conversion of UDP-Gal p to UDP-Gal f catalyzed by the flavoenzyme UDP-galactopyranose mutase (UGM). Because UGM is essential for the survival and proliferation of several pathogens, there is interest in understanding the catalytic mechanism to aid inhibitor development. Herein, we have used kinetic measurements and molecular dynamics simulations to explore the features of UGM that control the rate-limiting step (RLS). We show that UGM from the pathogenic fungus Aspergillus fumigatus also catalyzes the isomerization of UDP-arabinopyranose (UDP-Ara p), which differs from UDP-Gal p by lacking a -CH2-OH substituent at the C5 position of the hexose ring. Unexpectedly, the RLS changed from a chemical step for the natural substrate to product release with UDP-Ara p. This result implicated residues that contact the -CH2-OH of UDP-Gal p in controlling the mechanistic path. The mutation of one of these residues, Trp315, to Ala changed the RLS of the natural substrate to product release, similar to the wild-type enzyme with UDP-Ara p. Molecular dynamics simulations suggest that steric complementarity in the Michaelis complex is responsible for this distinct behavior. These results provide new insight into the UGM mechanism and, more generally, how steric factors in the enzyme active site control the free energy barriers along the reaction path.

Protein targets of thiazolidinone derivatives in Toxoplasma gondii and insights into their binding to ROP18.

BACKGROUND: Thiazolidinone derivatives show inhibitory activity (IC50) against the Toxoplasma gondii parasite, as well as high selectivity with high therapeutic index. To disclose the target proteins of the thiazolidinone core in this parasite, we explored in silico the active sites of different T. gondii proteins and estimated the binding-free energy of reported thiazolidinone molecules with inhibitory effect on invasion and replication of the parasite inside host cells. This enabled us to describe some of the most suitable structural characteristics to design a compound derived from the thiazolidinone core. RESULTS: The best binding affinity was observed in the active site of kinase proteins, we selected the active site of the T. gondii ROP18 kinase, because it is an important factor for the virulence and survival of the parasite. We present the possible effect of a derivative of thiazolidinone core in the active site of T. gondii ROP18 and described some characteristics of substituent groups that could improve the affinity and specificity of compounds derived from the thiazolidinone core against T. gondii. CONCLUSIONS: The results of our study suggest that compounds derived from the thiazolidinone core have a preference for protein kinases of T. gondii, being promising compounds for the development of new drugs with potential anti-toxoplasmosis activity. Our findings highlight the importance of use computational studies for the understanding of the action mechanism of compounds with biological activity.

Consistent Principal Component Modes from Molecular Dynamics Simulations of Proteins.

Principal component analysis is a technique widely used for studying the movements of proteins using data collected from molecular dynamics simulations. In spite of its extensive use, the technique has a serious drawback: equivalent simulations do not afford the same PC-modes. In this article, we show that concatenating equivalent trajectories and calculating the PC-modes from the concatenated one significantly enhances the reproducibility of the results. Moreover, the consistency of the modes can be systematically improved by adding more individual trajectories to the concatenated one.

In silico study of Moxifloxacin derivatives with possible antibacterial activity against a resistant form of DNA gyrase from Porphyromonas gingivalis.

We performed a homology modeling of the structure of a non-mutated and mutated Ser83→Phe DNA gyrase of Porphyromonas gingivalis. The model presented structural features conserved in type II topoisomerase proteins. We designed and evaluated in silico structural modifications to the core of Moxifloxacin by molecular docking, predicted toxicity and steered molecular dynamics simulations (SMD). Our results suggest that 8D derivative of Moxifloxacin could present a strong inhibitory activity in Porphyromonas gingivalis bacteria that exhibits resistance to some conventional fluoroquinolone drugs. Also, our results suggest that hydrophobic radicals in the hydroxyl group at position 3 of the quinolone core would increase the antibacterial activity of the compound when a reported mutation Ser83→Phe is present in the DNA gyrase protein. In addition, new candidates that could have a higher antibacterial activity compared to Moxifloxacin in non-resistant bacteria are proposed.

QM/MM molecular dynamics study of the galactopyranose → galactofuranose reaction catalysed by Trypanosoma cruzi UDP-galactopyranose mutase.

The enzyme UDP-Galactopyranose Mutase (UGM) catalyses the conversion of galactopyranose into galactofuranose. It is known to be critical for the survival and proliferation of several pathogenic agents, both prokaryotic and eukaryotic. Among them is Trypanosoma cruzi, the parasite responsible for Chagas' disease. Since the enzyme is not present in mammals, it appears as a promising target for the design of drugs to treat this illness. A precise knowledge of the mechanism of the catalysed reaction would be crucial to assist in such design. In this article we present a detailed study of all the putative steps of the mechanism. The study is based on QM/MM free energy calculations along properly selected reaction coordinates, and on the analysis of the main structural changes and interactions taking place at every step. The results are discussed in connection with the experimental evidence and previous theoretical studies.