Candidate Structure Prioritization in Non-Target Screening with Predicted Collision Cross-Sections and Uncertainty Quantification

Abstract Candidate structure prioritization in non-target screening (NTS) can be performed using collision cross-sections (CCSs) predicted by a machine-learning (ML) model, with a threshold on the relative difference between measured and in silico CCSs. Such thresholds generally are heuristic and lack a statistical foundation. In this context, we have constructed a conformal predictive system to quantify the uncertainty of the ML-predicted CCSs and provide guarantees on the minimum percentage of prioritized true structures among the candidate lists of the NTS features detected in a sample. The system is designed as a wrapper around the open-source ML model GraphCCS and can, in principle, be combined with any CCS model. Additionally, we propose the detection of the applicability domain of the conformal predictive system based on maximum common-edge subgraph (MCES) molecular distances. We showcase the application of our tool to two independent in silico NTS datasets. Results on the two tested external datasets suggest that applicability domain detection based on the MCES distance allows to maintain the guarantees on the minimum percentage of true structures prioritized. In terms of candidate structure prioritization performance, on average ∼16% of incorrect candidates could be deprioritized while retaining 90% of all true structures. Although this constitutes a modest 8% improvement over random prioritization, our tool additionally offers adaptivity of removal efficiency to the difficulty of the NTS candidates based on quantified uncertainty. Furthermore, the framework can potentially be combined with other analytical properties to improve candidate structure prioritization in NTS.