Dataset Information

Classifiers and their Metrics Quantified.

ABSTRACT: Molecular modeling frequently constructs classification models for the prediction of two-class entities, such as compound bio(in)activity, chemical property (non)existence, protein (non)interaction, and so forth. The models are evaluated using well known metrics such as accuracy or true positive rates. However, these frequently used metrics applied to retrospective and/or artificially generated prediction datasets can potentially overestimate true performance in actual prospective experiments. Here, we systematically consider metric value surface generation as a consequence of data balance, and propose the computation of an inverse cumulative distribution function taken over a metric surface. The proposed distribution analysis can aid in the selection of metrics when formulating study design. In addition to theoretical analyses, a practical example in chemogenomic virtual screening highlights the care required in metric selection and interpretation.

SUBMITTER: Brown JB

PROVIDER: S-EPMC5838539 | biostudies-literature | 2018 Jan

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Classifiers and their Metrics Quantified.

Brown J B JB

Molecular informatics 20180123 1-2

Molecular modeling frequently constructs classification models for the prediction of two-class entities, such as compound bio(in)activity, chemical property (non)existence, protein (non)interaction, and so forth. The models are evaluated using well known metrics such as accuracy or true positive rates. However, these frequently used metrics applied to retrospective and/or artificially generated prediction datasets can potentially overestimate true performance in actual prospective experiments. H ...[more]

PMID: 29360259

Similar Datasets

Project description:Formalin-fixed, paraffin-embedded (FFPE) tissues have many advantages for identification of risk biomarkers, including wide availability and potential for extended follow-up endpoints. However, RNA derived from archival FFPE samples has limited quality. Here we identified parameters that determine which FFPE samples have the potential for successful RNA extraction, library preparation, and generation of usable RNAseq data. We optimized library preparation protocols designed for use with FFPE samples using seven FFPE and Fresh Frozen replicate pairs, and tested optimized protocols using a study set of 130 FFPE biopsies from women with benign breast disease. Metrics from RNA extraction and preparation procedures were collected and compared with bioinformatics sequencing summary statistics. Finally, a decision tree model was built to learn the relationship between pre-sequencing lab metrics and qc pass/fail status as determined by bioinformatics metrics.. Samples that failed bioinformatics qc tended to have low median sample-wise correlation within the cohort (Spearman correlation < 0.75), low number of reads mapped to gene regions (< 25 million), or low number of detectable genes (11,400 # of detected genes with TPM > 4). The median RNA concentration and pre-capture library Qubit values for qc failed samples were 18.9 ng/ul and 2.08 ng/ul respectively, which were significantly lower than those of qc pass samples (40.8 ng/ul and 5.82 ng/ul). We built a decision tree model based on input RNA concentration, input library qubit values, and achieved an F score of 0.848 in predicting QC status (pass/fail) of FFPE samples. We provide a bioinformatics quality control recommendation for FFPE samples from breast tissue by evaluating bioinformatic and sample metrics. Our results suggest a minimum concentration of 25 ng/ul FFPE-extracted RNA for library preparation and 1.7 ng/ul pre-capture library output to achieve adequate RNA-seq data for downstream bioinformatics analysis.

Dataset Information

Classifiers and their Metrics Quantified.

Publications

Classifiers and their Metrics Quantified.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets