ATRPred

ATRPred predicts individual patient response to anti-tumor necrosis factor (anti-TNF) therapy in rheumatoid arthritis (RA) using plasma protein expression data.

Key Features:

• Target phenotype: Prediction of response to anti-TNF therapy in rheumatoid arthritis (RA).
• Data type: Plasma protein expression measured by proximity extension assays (PEA) using four panels of 92 proteins each.
• Sample cohort: Data were collected from 144 participants undergoing anti-TNF treatment, with 89 samples retained after quality control.
• Molecular stratification: Two molecular sub-groups (endotypes) were identified from plasma protein profiles, differing in gender distribution and disease activity but not predictive of treatment response.
• Outcome measure: Responders and non-responders were defined by change in Disease Activity Score (DAS) after six months of anti-TNF therapy.
• Machine learning: Classifier development employed a 5-fold nested cross-validation scheme to identify proteins associated with response.
• Biomarkers: Seventeen proteins were identified as significantly associated with treatment response.
• Performance: Reported classifier metrics are accuracy 81%, sensitivity 75%, and specificity 86%.
• Implementation: Implemented in R.

Scientific Applications:

• Predictive stratification: Predicts individual likelihood of positive response to anti-TNF therapy in RA patients.
• Biomarker identification: Identifies protein biomarkers associated with treatment response, including a 17-protein signature.
• Endotype characterization: Enables identification of molecular sub-groups (endotypes) from plasma proteomics for studies of disease heterogeneity.
• Translational research: Supports studies linking plasma protein profiles to clinical outcomes in RA.

Methodology:

Plasma protein expression was measured by proximity extension assays on four 92-protein panels; quality control produced 89 samples from an initial 144 participants; patients were labeled responder or non-responder by change in DAS after six months; a machine learning classifier was developed using a 5-fold nested cross-validation scheme, identifying 17 proteins associated with response; implementation is in R.