A comparison of model performance for six quantitative structure-activity relationship packages that predict acute toxicity to fish
Bibliographic record
Abstract
Some regulatory programs rely on quantitative structure-activity relationship (QSAR) models to predict toxic effects to biota. Many currently existing QSAR models can predict the effects of a wide range of substances to biota, particularly aquatic biota. The difficulty for regulatory programs is in choosing the appropriate QSAR model or models for application in their new and existing substances programs. We evaluated model performance of six QSAR modeling packages: Ecological Structure Activity Relationship (ECOSAR), TOPKAT, a Probabilistic Neural Network (PNN), a Computational Neural Network (CNN), the QSAR components of the Assessment Tools for the Evaluation of Risk (ASTER) system, and the Optimized Approach Based on Structural Indices Set (OASIS) system. Using a testing data set of 130 substances that had not been included in the training data sets of the QSAR models under consideration, we compared model predictions for 96-h median lethal concentrations (LC50s) to fathead minnows to the corresponding measured toxicity values available in the AQUIRE database. The testing data set was heavily weighted with neutral organics of low molecular weight and functionality. Many of the testing data set substances also had a nonpolar narcosis mode of action and/or were chlorinated. A variety of statistical measures (correlation coefficient, slope and intercept from a linear regression analysis, mean absolute and squared difference between log prediction and log measured toxicity, and the percentage of predictions within factors of 2, 5, 10, 100, and 1,000 of measured toxicity values) indicated that the PNN model had the best model performance for the full testing data set of 130 substances. The rank order of the remainder of the models depended on the statistical measure employed. TOPKAT also had excellent model performance for substances within its optimum prediction space. Only 37% of the substances in the testing data set, however, fell within this optimum prediction space.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".