Benchmarking of Machine Learning classifiers on plasma proteomic for COVID-19 severity prediction through interpretable artificial intelligence

APNet, an explainable sparse deep learning model to discover differentially active drivers of severe COVID-19

“…Some models exhibited eXplainable AI (XAI) features by deploying Shapley additive explanation (SHAP) values, the minimal-optimal variables method or a random forest explainer. In the same work, we managed to dovetail an explainable, computational pipeline to benchmark a wide assortment of ML tools on predicting COVID-19 severity from Olink plasma proteomics which revealed Multi-Layer Perceptron (MLP) as the highest-performing algorithm (Dimitsaki et al, 2023).…”

“…Some models exhibited eXplainable AI (XAI) features by deploying Shapley additive explanation (SHAP) values, the minimal-optimal variables method or a random forest explainer. In the same work, we managed to dovetail an explainable, computational pipeline to benchmark a wide assortment of ML tools on predicting COVID-19 severity from Olink plasma proteomics which revealed Multi-Layer Perceptron (MLP) as the highest-performing algorithm (Dimitsaki et al, 2023). However, most of the above studies can partially approximate proteomic non-linear dynamics (e.g., post-translational modifications, protein co-expression networks, complex formation, and subcellular localization), thus missing signaling proteins that may drive critical COVID-19 pathways.…”

“…However, most of the above studies can partially approximate proteomic non-linear dynamics (e.g., post-translational modifications, protein co-expression networks, complex formation, and subcellular localization), thus missing signaling proteins that may drive critical COVID-19 pathways. Moreover, these studies’ ML/DL findings often lack extensive external validation in large independent datasets, while their biological explainability is usually restricted to mere feature ranking (Paul et al, 2023) (Dimitsaki et al, 2023).…”

Plasma Proteins Associated with COVID-19 Severity in Puerto Rico

“…Patients with severe disease present a cytokine storm in the blood that may include elevation of TNF-α, CCL2, CXCL10 (IP-10), MCP-1, MIP-1α, GCSF, IL-1β, IL-2, IL-6, IL-7, IL-8, IL-10, and IL-17 [2][3][4][5][6][7][8][9]. A recent study using machine learning/artificial intelligence approaches in plasma proteomics datasets of COVID-19 patients linked severe disease with B cell dysfunction, increased inflammation, activation of Toll-like receptors, and decreased activation of developmental and immune mechanisms such as SCF/c-Kit signaling [10]. The proteins identified by artificial intelligence with the highest predictive values for COVID-19 disease severity were the following: CRK-like proto-oncogene, adaptor protein (CRKL), interleukin 1 receptor-associated kinase 1 (IRAK1), NF-kappa-B essential modulator/inhibitor of nuclear factor kappa-B kinase subunit gamma (NEMO/IKBKG), axis inhibition protein 1 (AXIN1), serine/arginine-rich protein-specific kinase 2 (SRPK2), and the cytoplasmic histidine-TRNA ligase (HARS1) [10].…”

Risk Factors and Prediction of 28-Day-All Cause Mortality Among Critically Ill Patients with Acute Pancreatitis Using Machine Learning Techniques: A Retrospective Analysis of Multi-Institutions

“… 19 At present, the SHAP analysis is widely used in the medical field, 20 , 21 and has also been applied as an interpretation approach for ML models. 22 , 23 …”