References
Everything this site is built on, in one place: the paper the method comes from, the work that revisits it, the data sources and the toolkit. If a number on this site is not traceable to something on this page, it should not be on this site.
The original method
Wessel MD, Jurs PC, Tolan JW, Muskal SM. Prediction of human intestinal absorption of drug compounds from molecular structure. Journal of Chemical Information and Computer Sciences 1998;38(4):726-735. doi:10.1021/ci980029a
86 compounds, 728 descriptors reduced to 127, a genetic algorithm choosing six for a 6-4-1 neural network, and reported errors of 9.40 training and 16.00 on an external set of ten. This is the model reproduced on How it works.
The current work
Muskal SM, Wessel MD. Human Intestinal Absorption Revisited: The 1998 Descriptors Survive, and the Method Extends to Permeability. bioRxiv, posted 11 October 2026, version 1. doi:10.64898/2026.10.03.756481
The reproduction of the 1998 model, the search over 1,301 modern descriptors and 780 compounds, the ceiling set by disagreement among published labels, the applicability flag, and the Caco-2 small-molecule and two cyclic peptide models. Every measured number quoted on this site comes from this work. The full text is on bioRxiv under CC BY 4.0.
Abstract
In 1998 we predicted percent human intestinal absorption for 86 drug and drug-like compounds from six molecular descriptors chosen by a genetic algorithm and a small neural network. That result rested on one division of the compounds into training, cross-validation and external prediction groups. Repeating the division shows it was a fortunate one. Across 250 stratified extractions, 14 match the published cross-validation error and only one matches the published external prediction error. An extraction is one draw from a wide distribution, so we repeated every experiment 250 times and report the result as a distribution. Reconstructed this way, the published six descriptors give a mean extraction error roughly twice the external prediction value we reported in 1998. We then applied the same six descriptors, unchanged, to a larger set of 780 absorption measurements from a public compilation. Training and extraction error come to within fifteen percent of one another, and under a random forest the six still reach R 0.63. Four datasets are modeled here: 780 small molecules with percent absorbed measured in man, 3,771 small molecules and 1,281 cyclic peptides measured across Caco-2 cell monolayers, and 7,298 cyclic peptides measured across artificial membranes. Each is its own prediction problem, and one procedure is applied to all four. A genetic algorithm searches descriptor sets whose size is free to vary, scoring each candidate by the out-of-bag error of a random forest fitted inside that extraction's training group. The descriptors chosen most often across the 250 extractions form one consensus set for each dataset, and that set is what we deploy. The four models are served at predicthia.ai, where a structure can be drawn or pasted, scored against its training set for similarity, and the training data with fitted models downloaded. Each model served is its fixed consensus set fitted on every compound in its dataset; the accuracy quoted beside a prediction comes from evaluating that same set across 250 extractions. The fixed sets hold 25 descriptors for absorption, 55 for Caco-2 small-molecule permeability, 50 for Caco-2 peptide permeability and 25 for artificial-membrane permeability.
Keywords: human intestinal absorption; QSPR; genetic algorithm; descriptor selection; cyclic peptides; membrane permeability; oral bioavailability; repeated extraction
Data sources
The human intestinal absorption and bioavailability datasets compiled by the Institute of Biophysics and Biomedical Engineering, Bulgarian Academy of Sciences, for the COSMOS project. biomed.bas.bg/qsarmm/Human_intestinal_absorption_and_bioavalability
The source of every absorption label on this site: 783 compounds in the compilation, two entries for a fluorescent paracellular marker excluded, and one compound whose geometry would not embed, leaving the 780 the model is fitted to. Where a compound carries several published values, the median is used and the spread is reported rather than discarded. The per-compound provenance travels with the table on Download.
OpenADMET-ExpansionRx challenge, full post-challenge release. Lead optimization measurements contributed by Expansion Therapeutics and released by the OpenADMET consortium under CC-BY-4.0. huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data and openadmet.org/datasetsmodels
The source of the 3,771 small molecules behind the Caco-2 small-molecule model.
CycPeptMPDB, the Cyclic Peptide Membrane Permeability Database. cycpeptmpdb.com
The source of both peptide models: 1,281 peptides measured on Caco-2 monolayers and 7,298 measured on artificial membranes of the PAMPA class. The two are different assays on different populations and are kept apart throughout, here and in the service.
Toolkit
RDKit: open source cheminformatics. rdkit.org
Every descriptor on this site is computed with RDKit, including the reimplementations of the six 1998 descriptors, which were rebuilt from their written definitions because the software that computed them in 1998 no longer runs. The exact pinned version travels with each prediction and is reported by the service.
Ketcher, the structure editor embedded on Try it, vendored and served from this site rather than loaded from a third party.
How to cite
For the original method, cite the 1998 paper. For anything computed on this site, cite the 2026 paper and say which model you used, because the two peptide models are separate and a result from one is not a result from the other.
Wessel MD, Jurs PC, Tolan JW, Muskal SM. Prediction of human intestinal absorption of drug compounds from molecular structure. J Chem Inf Comput Sci. 1998;38(4):726-735. doi:10.1021/ci980029a
Muskal SM, Wessel MD. Human Intestinal Absorption Revisited: The 1998 Descriptors Survive, and the Method Extends to Permeability. bioRxiv 2026. doi:10.64898/2026.10.03.756481. Predictions computed with predicthia, https://predicthia.ai/
If you quote a prediction, quote its interval and its applicability band with it. A value from this site without its interval is not a result this site produced.