Handcrafted versus non-handcrafted (self-supervised) features for the classification of antimicrobial peptides: complementary or redundant?

Abstract

Antimicrobial peptides (AMPs) have received a great deal of attention given their potential to become a plausible option to fight multi-drug resistant bacteria as well as other pathogens. Quantitative sequence-activity models (QSAMs) have been helpful to discover new AMPs because they allow to explore a large universe of peptide sequences and help reduce the number of wet lab experiments. A main aspect in the building of QSAMs based on shallow learning is to determine an optimal set of protein descriptors (features) required to discriminate between sequences with different antimicrobial activities. These features are generally handcrafted from peptide sequence datasets that are labeled with specific antimicrobial activities. However, recent developments have shown that unsupervised approaches can be used to determine features that outperform human-engineered (handcrafted) features. Thus, knowing which of these two approaches contribute to a better classification of AMPs, it is a fundamental question in order to design more accurate models. Here, we present a systematic and rigorous study to compare both types of features. Experimental outcomes show that non-handcrafted features lead to achieve better performances than handcrafted features. However, the experiments also prove that an improvement in performance is achieved when both types of features are merged. A relevance analysis reveals that non-handcrafted features have higher information content than handcrafted features, while an interaction-based importance analysis reveals that handcrafted features are more important. These findings suggest that there is complementarity between both types of features. Comparisons regarding state-of-the-art deep models show that shallow models yield better performances both when fed with non-handcrafted features alone and when fed with non-handcrafted and handcrafted features together.

Examining evolutionary scale modeling‐derived different‐dimensional embeddings in the antimicrobial peptide classification through a KNIME workflow

“…These results confirm the results reported in (Garcia‐Jacas et al, 2022; García‐Jacas et al, 2022), where was demonstrated that non‐DL based models achieve comparable‐to‐superior performances to DL‐based models in the prediction of AMPs when methodologically principled studies are performed. Several works justifying that DL‐based models improve the AMP classification, often by narrow performance margins, present modeling biases related to the quality and diversity of the sequence‐based protein features, lack of feature selection approaches, and poor use of ensemble classifiers.…”

“…In this way, the burden of a-priori feature engineering process is removed. But, as it has been demonstrated elsewhere (Garcia-Jacas et al, 2022;García-Jacas et al, 2022;Muratov et al, 2020), DL architectures lead to build QSAR models with comparable-to-inferior performances regarding non-DL based QSAR models when using small-and medium-sized sets. This is because such datasets are not large enough (Manibardo et al, 2021;Oyedare & Park, 2019) to automatically get learned (nonhandcrafted) features with better modeling abilities than calculated (handcrafted) features that can be selected from them (Garcia-Jacas et al, 2022;García-Jacas et al, 2022).…”

Cell‐penetrating peptides predictors: A comparative analysis of methods and datasets

“…Considering that there are fewer CPPs than AMPs the ideas of transfer learning [45] should be explored. A recent work has also produced a self-learned embedding [46] based on 250 million protein sequences that has also shown its potential in AMP classification [47]. Also, with the advent of AlphaFold2 [48], ESMFold [49], and Roset-taFold [50], structure information could be included to assess if this information can improve classification performance.…”

Analysis, Modeling, and Target-Specific Predictions of Linear Peptides Inhibiting Virus Entry

“…The existence of natural AVPs suggests that synthetic peptides may also have antiviral potential, but random sequence generation for AVP screening is not a cost-effective approach. Quantitative sequence-activity models (QSAMs) have been proved useful for the discovery of new AMPs and AVPs in particular, as they allow the exploration of a large range of peptide sequences and help reduce the number of laboratory experiments , …”

“…3 The existence of natural AVPs suggests that synthetic peptides may also have antiviral potential, but random sequence generation for AVP screening is not a cost-effective approach. Quantitative sequence-activity models (QSAMs) have been proved useful for the discovery of new AMPs and AVPs in particular, as they allow the exploration of a large range of peptide sequences and help reduce the number of laboratory experiments 4,5 Several papers have discussed the development of artificial peptide sequences with AVP activity. Thakur et al 6 proposed an AVP prediction algorithm based on a model derived from experimentally validated positive and negative data sets and included the model into the web tool AVPpred.…”