DeepDigest: Prediction of Protein Proteolytic Digestion with Deep Learning

Abstract

In shotgun proteomics, it is essential to accurately determine the proteolytic products of each protein in the sample for subsequent identification and quantification, because these proteolytic products are usually taken as the surrogates of their parent proteins in the further data analysis. However, systematical studies about the commonly used proteases in proteomics research are insufficient, and there is a lack of easy-to-use tools to predict the digestibilities of these proteolytic products. Here, we propose a novel sequence-based deep learning model -DeepDigest, which integrates convolutional neural networks and long-short term memory networks for digestibility prediction of peptides. DeepDigest can predict the proteolytic cleavage sites for eight popular proteases including trypsin, ArgC, chymotrypsin, GluC, LysC, AspN, LysN and LysargiNase. Compared with traditional machine learning algorithms, DeepDigest showed superior performance for all the eight proteases on a variety of datasets. Besides, some interesting characteristics of different proteases were revealed and discussed.

Investigation and Highly Accurate Prediction of Missed Tryptic Cleavages by Deep Learning

“…We compared the performance of dpMC with known tools for cleavage prediction such as DeepDigest, SVM, and information theory approaches within MC:pred and PeptideCutter () from ExPASy (Figure , Figure S5, and Supplementary Note). All the abovementioned tools, including dpMC, were tested on the same corresponding holdout data set of the six data sets (Table ).…”

“…So far, the probability of cleavage for tryptic peptide prediction has been based on the Keil rules , that describe a blockage of digestion when arginine or lysine is followed by proline and a reduction of cleavage when acidic amino acids flank either side of the corresponding arginine and lysine. However, such fixed rules cannot fully explain all experimentally observed missed cleavages, leading to a flurry of approaches to better explain and predict missed cleavages. ,− Moreover, as trypsin cleavage is essentially a probabilistic event, no fixed rules could fully explain which peptide bonds will be cleaved and which ones will not be cleaved. The oversimplified cleavage rules applied in experimental data analysis can often lead to false or inaccurate identifications and quantifications based on the peptide spectrum matches (PSMs). , Accurate annotations of missed tryptic cleavages will remove unlikely sequences and lower the complexity of the database, which, in turn, will result in increased sensitivity and specificity, while decreasing the analysis time. ,, …”

Proteome-pI2.0: proteome isoelectric point database update

“…Future plans in this respect include adding the prediction of peptides’ hydrophobicity, retention time ( 49 ), electrophoretic mobility ( 50 ), and the use of more sophisticated methods than can be utilized for the prediction of in silico digests [e.g. DeepDigest ( 51 )]. Finally, adding information about the uniqueness of peptides versus coverage after digestion would be also valuable.…”

Evaluation of an integrative Bayesian peptide detection approach on a combinatorial peptide library

“…As peptides of more abundant proteins have a higher representation in a sample, this can be reflected in their elevated prior probabilities by using protein abundance databases such as PaxDb 54 . One could also incorporate the probabilistic enzymatic cleavage behavior by precomputing cleavage sites on the reference protein database using a recent deep-learning algorithm DeepDigest for the purpose 55 . Although we did not consider PTMs in our study, there are no principal obstacles in incorporating them, which again allows assigning different prior probabilities to individual PTMs, and raise those that were already evidenced, e.g., in the UniProt database 56 .…”