Explore the vast range of titles available.

Peptides play important roles in regulating biological processes and form the basis of a multiplicity of therapeutic drugs. To date, only about 300 peptides in human have confirmed bioactivity, although tens of thousands have been reported in the literature. The majority of these are inactive degradation products of endogenous proteins and peptides, presenting a needle-in-a-haystack problem of identifying the most promising candidate peptides from large-scale peptidomics experiments to test for bioactivity. To address this challenge, we conducted a comprehensive analysis of the mammalian peptidome across seven tissues in four different mouse strains and used the data to train a machine learning model that predicts hundreds of peptide candidates based on patterns in the mass spectrometry data. We provide in silico validation examples and experimental confirmation of bioactivity for two peptides, demonstrating the utility of this resource for discovering lead peptides for further characterization and therapeutic development. Bioactive peptides regulate many physiological functions but progress in discovering them has been slow. Here, the authors use a machine learning framework to predict mammalian peptide candidates from the global and local structure of large-scale tissue-specific mass spectrometry data.

Tandem mass-spectrometry has become the method of choice for high-throughput, quantitative analysis in proteomics. However, since the link between the peptides and the proteins they originate from is typically broken, identification of the analyzed peptides relies on matching of the fragmentation spectra (MS2) to theoretical spectra of possible candidate peptides, often filtered for precursor ion mass. To this end, peptide-spectrum matching algorithms score the concordance between the experimental and the theoretical spectra of candidate peptides, by evaluating the number (or proportion) of theoretically possible fragment ions observed in the experimental spectra, without any discrimination. However, the assumption that each theoretical fragment is just as likely to be observed is inaccurate. On the contrary, MS2 spectra often have few dominant fragments. We propose a novel prediction algorithm based on a hidden Markov model, which allow for the training process to be carried out very efficiently. Using millions of MS/MS spectra generated in our lab, we found an overall good reproducibility across different fragmentation spectra, given the precursor peptide and charge state. This result implies that there is indeed a pattern to fragmentation that justifies using machine learning methods. Furthermore, the overall agreement between spectra of the same peptide at the same charge state serves as an upper limit on how well prediction algorithms can be expected to perform. We have investigated the performance of a third order HMM model, trained on several million MS2 spectra, in various ways. Compared to a mock model, in which the fragment ions and their intensities are shuffled, we see a clear difference in prediction accuracy using our model. This result indicates that our model can pick up meaningful patterns, i.e. we can indeed learn the fragmentation process. Secondly, looking at the variability of the prediction performance by varying the train/test data split, in a K-fold cross validation scheme, we observed an overall robust model that performs well independent of the specific peptides that are present in the training data. Last but not least, we propose that the real value of this model is as a pre-processing step in the peptide identification process, by discerning fragment ions that are unlikely to be intense for a given candidate peptide, rather than using the actual predicted intensities. As such, probabilistic measures of concordance between experimental and theoretical spectra, would leverage better statistics.

Many secreted endogenous peptides rely on signalling pathways to exert their function in the body. While peptides can be discovered through high throughput technologies, their cognate receptors typically cannot, hindering the understanding of their mode of action. We investigate the use of AlphaFold-Multimer for identifying the cognate receptors of secreted endogenous peptides in human receptor libraries without any prior knowledge about likely candidates. We find that AlphaFold's predicted confidence metrics have strong performance for prioritizing true peptide-receptor interactions. By applying transmembrane topology prediction using DeepTMHMM, we further improve performance by detecting and filtering biologically implausible predicted interactions. In a library of 1112 human receptors, the method ranks true receptors in the top percentile on average for 11 benchmark peptide-receptor pairs. Competing Interest Statement The authors have declared no competing interest.