Gene expression and response prediction to amisulpride in the OPTiMiSE first episode psychoses.

A fundamental shortcoming in the current treatment of schizophrenia is the lack of valid criteria to predict who will respond to antipsychotic treatment. The identification of blood-based biological markers of the therapeutic response would enable clinicians to identify the subgroup of patients in whom conventional antipsychotic treatment is ineffective and offer alternative treatments. As part of the Optimisation of Treatment and Management of Schizophrenia in Europe (OPTiMiSE) programme, we conducted an RNA-Seq analysis on 188 subjects with first episode psychosis, all of whom were subsequently treated with amisulpride for 4 weeks. We compared gene expression on total RNA from patients' blood before and after treatment and identified 32 genes for which the expression changed after treatment in good responders only. These findings were replicated in an independent sample of 24 patients with first episode psychosis. Six genes showed a significant difference in expression level between good and poor responders before starting treatment, allowing to predict treatment outcome with a predictive value of 93.8% when combined with clinical features. Collectively, these findings identified new mechanisms to explain symptom improvement after amisulpride medication and highlight the potential of combining gene expression profiling with clinical data to predict treatment response in first episode psychoses.

Correction: Stratification and prediction of remission in first-episode psychosis patients: the OPTiMiSE cohort study.

The original Article did not feature the list of collaborators. This has now been corrected in the PDF and HTML versions of this Article.

Gentle Nearest Neighbors Boosting over Proper Scoring Rules.

Tailoring nearest neighbors algorithms to boosting is an important problem. Recent papers study an approach, UNN, which provably minimizes particular convex surrogates under weak assumptions. However, numerical issues make it necessary to experimentally tweak parts of the UNN algorithm, at the possible expense of the algorithm's convergence and performance. In this paper, we propose a lightweight Newton-Raphson alternative optimizing proper scoring rules from a very broad set, and establish formal convergence rates under the boosting framework that compete with those known for UNN. To the best of our knowledge, no such boosting-compliant convergence rates were previously known in the popular Gentle Adaboost's lineage. We provide experiments on a dozen domains, including Caltech and SUN computer vision databases, comparing our approach to major families including support vector machines, (Ada)boosting and stochastic gradient descent. They support three major conclusions: (i) GNNB significantly outperforms UNN, in terms of convergence rate and quality of the outputs, (ii) GNNB performs on par with or better than computationally intensive large margin approaches, (iii) on large domains that rule out those latter approaches for computational reasons, GNNB provides a simple and competitive contender to stochastic gradient descent. Experiments include a divide-and-conquer improvement of GNNB exploiting the link with proper scoring rules optimization.