Poster Presentation AUS-oMicS 2025

ScavenPred: A Random Forest Antioxidant Peptide Predictor Built on Amino Acid Descriptors (119847)

Ken KUA Agbo 1 2 , Afaq AS Shah 3 , Utpal UB Bose 1 2 , Michelle MC Colgrave 1 2 , Angela AJ Juhasz 1
  1. Australian Research Council Centre of Excellence for Innovations in Peptide and Protein Science, School of Science, Edith Cowan University, Joondalup, Perth, Western Australia, Australia
  2. CSIRO Agriculture and Food, 306 Carmody Rd, St Lucia, QLD 4067, Queensland, Australia
  3. Centre for AI and Machine Learning, School of Science, Edith Cowan University, 270 Joondalup Dr, Joondalup, Perth, Western Australia, Australia

Identifying candidate bioactive peptides (BPs) from food products for nutraceutical applications is laborious and challenging using the typical liquid chromatography-mass spectrometry (LC-MS)-based proteomics workflow. Thus, the in-silico screening approach can offer a solution for the initial screening to reduce the number of candidate peptides for in-vitro activity testing. Various machine learning-based tools have been developed previously for activity screening, however, the lack of negative datasets and inadequate selection of peptide physicochemical features for model training leads to poor prediction outcomes, limiting the use of in-silico methods. To address this challenge, we developed novel protocols for peptide feature selection and negative dataset generation. We developed a novel antioxidant predictor, Scavenpred and collected MS data to predict the antioxidant properties of the identified peptides. We also benchmarked our tool against the two most popular antioxidant peptide predictors, AnoxPP and Anoxpepred using experimentally validated antioxidant peptide sets. We observed accuracy and MCC scores of 0.87 & 0.66 for ScavenPred, which outperformed AnoxPP and Anoxpepred with accuracies and MCC scores of 0.79 & 0.6 and 0.42 & 0.19 respectively. We collected Data-Dependent Acquisition (DDA) data from urea and ethanol solvents extracted and trypsin-digested faba bean and sorghum grain samples and we detected 1,373 faba bean and sorghum 523 peptides at 1% global FDR). Of these identified peptides Scavenpred predicted 74 (5.46%) faba bean and 39 (7.46%) sorghum peptides with high antioxidant potential (threshold 0.75). Using the same faba bean and sorghum peptide sets with AnoxPP (threshold 0.75) we identified 1,244/1,371 (90.74%) faba bean and 456/481 (94.80%) sorghum peptides, while Anoxpepred predicted (threshold 0.5), 119/1,371 (8.65%) faba bean and 62/481 (13%) sorghum peptides with antioxidant potential, respectively. Future studies will be conducted to synthesise the candidate peptides and measure their antioxidant properties using in-vitro assays for free radical scavenging activity such as DPPH and ABTS.