Cheng Chang

This paper introduces π-SPECFormer, a foundation model for mass spectrometry-based proteomics that decodes the dark proteome by pre-training on unannotated spectra and outperforms baselines in peptide sequencing and modification detection.

The paper introduces π-MSNet, a large-scale proteomics dataset for AI training and benchmarking, showing performance improvements when models are retrained on it.

π-MNovo improves de novo peptide sequencing for metaproteomics by adapting to microbial spectra and selecting candidates based on evidence, achieving higher recall than existing models.

This paper introduces a method to control false discovery rates in deep-learning-based de novo peptide sequencing, improving reliability of peptide identifications.

The paper introduces πDIA-CLIP, a zero-shot framework for DIA proteomics data analysis that improves protein identification and reduces false positives across diverse datasets.

MolSpecFlow bridges the gap between molecules and mass spectra using a hybrid flow matching model, achieving state-of-the-art results in de novo generation and spectral simulation.

This paper introduces SRPS, a subtyping algorithm that improves classification accuracy and prognostic discrimination in multicentric proteomic studies, identifying PPIC as a significant biomarker in HCC.

π-PrimeNovo is a non-autoregressive deep learning model for de novo peptide sequencing that achieves higher accuracy and 69x faster performance than existing methods.

The paper trains machine learning models on multi-omics UK Biobank data to predict disease risk, finding metabolomics provides limited improvement over standard features but can replace them when unavailable.

The paper introduces Cell Decoder, a deep learning model that provides multi-scale interpretability for cell type decoding and data integration, showing its effectiveness in human and mouse datasets.

DeepRTAlign improves retention time alignment for large cohort LC-MS data, enhancing identification sensitivity without compromising quantitative accuracy and enabling predictive modeling for hepatocellular carcinoma recurrence.

The paper introduces π-HelixNovo, a Transformer-based model for de novo peptide sequencing that outperforms existing methods and improves taxonomic resolution in gut metaproteomes.

EnrichMiner provides a web server to help researchers explore and refine functional enrichment analysis results by removing redundancy and selecting representative terms.

DeepDigest predicts proteolytic cleavage sites for eight proteases using a deep learning model, outperforming traditional methods across multiple datasets.

The paper introduces AP3, an algorithm that improves proteotypic peptide prediction by incorporating peptide digestion probability, achieving higher accuracy than existing tools.

LFAQ improves label-free absolute protein quantification by correcting MS intensity biases using predicted peptide quantitative factors, achieving higher accuracy and precision, especially for low-abundance proteins.

PANDA is a proteomics data quantification tool that offers higher accuracy and precision than MaxQuant with reduced computation time, supporting both label-free and labeled quantification.

This study compares the effects of tap water, bottled purified water, and bottled natural water on developmental behaviors, lipid metabolism, and antioxidant capacity in juvenile zebrafish.