Scientists from Henan University in Kaifeng, China created a new prediction model for classifying nonclassical secreted proteins based on deep learning, publishing their findings in the Journal of Chemometrics (1). The scientists involved in the study include Xinhong Zhang and Yiru He from the Henan University School of Software, Binjie Wang and Fan Zhang from the university’s Radiological Department, and Chaoyang Liu of the Henan University School of Computer and Information Engineering.
Deep learning is a subset of machine learning that uses multi-layered neural networks, otherwise called deep neural networks (DNNs), to simulate the human brain’s complex decision-making power (2). Unlike machine learning, which needs to pre-process and organize unstructured data, deep learning eliminates some of the pre-processing that is usually involved with machine learning. While DNNs must have at minimum three or more layers, most DNNs have many more layers. DNNs are trained on large amounts of data to identify and classify phenomena, recognize patterns and relationships, evaluate possibilities, and make predictions and decisions. These additional layers can help refine and optimize predictions and decisions for greater accuracy. Deep learning can drive many applications and services that improve automation, performing analytical and physical tasks without human intervention and enabling everyday products and services. and enabling everyday products and services.
Read More: Deep Learning SERS in the Rapid Detection of Kidney Allograft Injuries
In this study, the scientists proposed an end-to-end nonclassical secreted protein prediction model based on deep learning, named DeepNCSPP. This model employs protein sequence information and sequence statistics information as input to predict whether it is a nonclassical secreted protein. Protein sequence information is extracted using bidirectional long- and short-term memory, while sequence statistics information is extracted using convolutional neural networks.
Accurate protein structure predictions, which have been enabled by advancements in machine learning algorithms, allow for entry points into probing structural mechanisms and integrating and querying different types of biochemical and biophysical results (3). Many nonclassical protein prediction methods that are used today involve manual feature selection. This type of process involves constructing sample features based on the physicochemical properties of proteins and position-specific scoring matrix (PSSM). However, these tasks can require researchers to perform tedious search work to obtain the physicochemical properties of proteins.
Read More: Using Thermostable Raman Interaction Profiling (TRIP) For Protein Binding Screening
Among the experiments conducted on the independent test data set, DeepNCSPP achieved excellent results, achieving an accuracy of 88.24%, a Matthews coefficient (MCC) of 77.01%, and an F1-score of 87.50%. Along with independent test data set testing and 10-fold cross-validation, these findings show that DeepNCSPP can achieve competitive performance with state-of-the art methods, while also being used as a reliable nonclassical secreted protein prediction model. A web server has been created for researchers’ convenience, with the link being found here: https://www.deepncspp.top/.
Artificial intelligence (AI) and machine learning are becoming more common in analytical science. For example, researchers from Tsinghua University and Beihang University in Beijing developed a deep-learning-based data processing framework that significantly improves the accuracy of dual-comb absorption spectroscopy (DCAS) in gas quantification analysis, Spectroscopy previously reported (4). By using a U-net model for etalon removal and a modified U-net combined with traditional methods for baseline extraction, their framework achieves high-fidelity absorbance spectra, even in challenging conditions with complex baselines and etalon effects. U-Net combined with adaptive iteratively reweighted penalized least squares (airPLS) is a hybrid approach for processing complex spectra data, tackling issues like etalon effects and complex baselines.
(1) Zhang, F.; Liu, C.; Wang, B.; et al. A Prediction Model of Nonclassical Secreted Protein Based on Deep Learning. J. Chemom. 2024, e3553. DOI: 10.1002/cem.3553
(2) What is Deep Learning? IBM 2024. https://www.ibm.com/topics/deep-learning (accessed 2024-5-22)
(3) Protein Structure Prediction. https://www.sciencedirect.com/topics/biochemistry-genetics-and-molecular-biology/protein-structure-prediction (accessed 2024-5-22)
(4) Workman Jr, J. Deep Learning Advances Gas Quantification Analysis in Near-Infrared Dual-Comb Spectroscopy. Spectroscopy 2024. https://www.spectroscopyonline.com/view/deep-learning-advances-gas-quantification-analysis-in-near-infrared-dual-comb-spectroscopy (accessed 2024-5-23)
Emerging Leader Highlights Innovations in Machine Learning, Chemometrics at SciX Awards Session
October 23rd 2024Five invited speakers joined Joseph Smith, the 2024 Emerging Leader in Molecular Spectroscopy, on stage to speak about trends in hyperspectral imaging, FT-IR, surface enhanced Raman spectroscopy (SERS), and more during the conference in Raleigh.
Advancing Forensic Science with Chemometrics: New Tools for Objective Evidence Analysis
October 22nd 2024A review by researchers from Curtin University comprehensively explores how chemometrics can revolutionize forensic science by offering objective and statistically validated methods to interpret evidence. The chemometrics approach seeks to enhance the accuracy and reliability of forensic analyses, mitigating human bias and improving courtroom confidence in forensic conclusions.