ExamPle: explainable deep learning framework for the prediction of plant small secreted peptides

Abstract

Motivation Plant Small Secreted Peptides (SSPs) play an important role in plant growth, development, and plant-microbe interactions. Therefore, the identification of SSPs is essential for revealing the functional mechanisms. Over the last few decades, machine learning-based methods have been developed, accelerating the discovery of SSPs to some extent. However, existing methods highly depend on handcrafted feature engineering, which easily ignores the latent feature representations and impacts the predictive performance. Results Here, we propose ExamPle, a novel deep learning model using Siamese network and multi-view representation for the explainable prediction of the plant SSPs. Benchmarking comparison results show that our ExamPle performs significantly better than existing methods in the prediction of plant SSPs. Also, our model shows excellent feature extraction ability by using dimension reduction tools. Importantly, by utilizing in silico mutagenesis (ISM) experiments, ExamPle can discover sequence characteristics and identify the contribution of each amino acid. The key novel principle learned by our model is that the head region of the peptide and some specific sequential patterns are strongly associated with the SSPs’ functions. Thus, ExamPle is a competitive model and tool for predicting plant SSPs and designing effective plant SSPs. Availability Our codes and datasets are available at https://github.com/Johnsunnn/ExamPle. Supplementary information Supplementary data are available at Bioinformatics online.

An Effective Plant Small Secretory Peptide Recognition Model Based on Feature Correction Strategy

“…These secondary structure data were then integrated into the SSP data. For the specific data processing process, please refer to ref , and this work constructs these data.…”

“…To alleviate the problem of false positive rate, Wei et al designed an SSP recognition framework named ExamPle. 35 The model encodes the protein sequence and secondary structure based on the transformer encoder and then uses a contrastive learning strategy to train the SSP predictor. This model not only solves the shortcomings of traditional manual feature extraction, it also uses the attention mechanism of transformer to explore the importance of subsequences through feature weights.…”

“…However, the software also has obvious flaws, and its prediction results usually have more false positive samples. To alleviate the problem of false positive rate, Wei et al designed an SSP recognition framework named ExamPle . The model encodes the protein sequence and secondary structure based on the transformer encoder and then uses a contrastive learning strategy to train the SSP predictor.…”

Improved super-resolution ribosome profiling reveals prevalent translation of upstream ORFs and small ORFs in Arabidopsis

“…Bioinformatic approaches based on evolutionary conservation or sequence homology to known sORFs have been attempted ( Olsen et al 2002 ; Lease and Walker 2006 ; Hanada et al 2007 , 2009 ; Zhou et al 2013 ; de Bang et al 2017 ; Feng et al 2023 ; Li et al 2023 ). However, accumulating evidence suggests that many sORFs have evolved recently during evolution ( Ruiz-Orera et al 2014 ; Sandmann et al 2023 ), and some sORFs are only conserved in specific families or small groups of plants.…”

Improved Super-Resolution Ribosome Profiling Revealed Prevalent Translation of Upstream ORFs and Small ORFs in Arabidopsis

“…Bioinformatic approaches based on evolutionary conservation or sequence homology to known sORFs have been attempted (Olsen et al, 2002;Lease and Walker, 2006;Hanada et al, 2007Hanada et al, , 2009Zhou et al, 2013;de Bang et al, 2017;Feng et al, 2023;Li et al, 2023). However, accumulating evidence suggests that many sORFs have evolved recently during evolution (Ruiz-Orera et al, 2014;Sandmann et al, 2023), and some sORFs are only conserved in specific families or small groups of plants.…”