Optimizing Artificial Neural Network Models for Metabolomics and Systems Biology: an Example Using HPLC Retention Index Data

Abstract

Background Artificial Neural Networks (ANN) are extensively used to model ‘omics’ data. Different modeling methodologies and combinations of adjustable parameters influence model performance and complicate model optimization. Methodology We evaluated optimization of four ANN modeling parameters (learning rate annealing, stopping criteria, data split method, network architecture) using retention index (RI) data for 390 compounds. Models were assessed by independent validation (I-Val) using newly measured RI values for 1492 compounds. Conclusion The best model demonstrated an I-Val standard error of 55 RI units and was built using a Ward’s clustering data split and a minimally nonlinear network architecture. Use of validation statistics for stopping and final model selection resulted in better independent validation performance than the use of test set statistics.

Challenges in Identifying the Dark Molecules of Life

“…The generalization of such predictive models depends on the application domain investigated, because using overly restricted domains for model building also leads to poor prediction in independent test sets (38). Molecular descriptors (39–42) also play a significant role in defining such application domains (38, 43, 44), as well as in defining the multiple algorithmic options and combinations of adjustable parameters involved in QSRR model building (42, 45). In general, RT factors (RF in the equation below, where T 0 represents the chromatographic column void time) are calculated to allow comparison between different chromatographic systems (32, 36).…”

“…In general, RT factors (RF in the equation below, where T 0 represents the chromatographic column void time) are calculated to allow comparison between different chromatographic systems (32, 36). Other studies utilize retention indices (28, 29), which are measures of relative RT based on reference compounds that elute immediately prior to and immediately following the analyte of interest (30, 45–48). RF=RT−T0T0. Cao et al (31), for example, conducted QSRR modeling based on theoretical molecular descriptors and experimental RTs of 93 authentic compounds analyzed with HILIC LC-MS. A predictive QSRR model based on a random forest algorithm achieved high predictive accuracy, with mean and median absolute errors of 0.52 min and 0.34 min (5.1% and 3.2%), respectively.…”

Predicting RP-LC retention indices of structurally unknown chemicals from mass spectrometry data

“…The simultaneous analysis of the samples and the r i calibrants under specified conditions has enabled the measurement of r i s of structurally unknown chemicals. 59,60 These measured r i s are then compared to the r i databases of structurally known chemicals to further increase the associated confidence in the generated identifications. [59][60][61] Additionally, recent studies have highlighted the use of quantitative structure retention relationship (QSRR) models to predict and populate the r i databases, employing molecular descriptors.…”

“…59,60 These measured r i s are then compared to the r i databases of structurally known chemicals to further increase the associated confidence in the generated identifications. [59][60][61] Additionally, recent studies have highlighted the use of quantitative structure retention relationship (QSRR) models to predict and populate the r i databases, employing molecular descriptors. [59][60][61][62] The QSRR methods for the prediction of r i values of structurally known chemicals have been a complementary strategy to the experimentally defined r i values.…”

Using LC Retention Times in Organic Structure Determination: Drug Metabolite Identification

“…One obstacle in RT prediction is that, unlike past experiences with gas chromatography [18], RTs for reverse phase LC tend to be less reproducible due to subtle effects of pH, underivatized silanols and mobile phase compositions; however, the robustness of RT assessment can be improved with RT indexing, for example using nitroalkanes [3], or amides [8] or drugs [11]. Valko’s Chromatographic Hydrophobicity Index (CHI) [19] is attractive and uses language familiar to drug metabolism scientists.…”