Sampling Strategy: An Overlooked Factor Affecting Artificial Intelligence Prediction Accuracy of Peptides’ Physicochemical Properties
Year: 2025
Type: article
Abstract: Accurate artificial intelligence (AI) prediction of peptide physicochemical properties is essential for advancing peptide‐based biomedicine, biotechnology, and bioengineering. However, the predictive performance of AI models is significantly affected by the representativeness of the training data. This study addresses the challenge of determining the optimal sample size and sampling methods to enhance the predictive accuracy and generalization capacity of AI models for estimating the aggregation propensity, hydrophilicity (logP), and isoelectric point (pI) of tetrapeptides in the complete sequence space. Four sampling methods are evaluated: Latin hypercube sampling, uniform design sampling, simple random sampling, and probability‐proportional‐to‐size sampling, across sample sizes ranging from 100 to 20 000. Our results highlight the critical role of sample size rather than the specific sampling method in achieving reliable and optimal predictive performance. This study provides valuable insights into the interplay between sample size, sampling strategies, and model performance, offering a foundational framework for optimizing data collection and AI model training for the prediction of peptides’ physicochemical properties, especially for prediction in the complete sequence space of longer peptides with more than four amino acids or mixed peptide systems, where the complexity of the sequence space increases exponentially.
Source: Advanced Intelligent Discovery
Authors: Meiru Yan, Ankeer Abuduhebaier, Haojin Zhou, Jiaqi Wang
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: diamond
Awards: G3928359561, G7441118358
OpenAlex ID: W4417290157