Prediction of drug absorption using multivariate statistics.

Literature data on compounds both well- and poorly-absorbed in humans were used to build a statistical pattern recognition model of passive intestinal absorption. Robust outlier detection was utilized to analyze the well-absorbed compounds, some of which were intermingled with the poorly-absorbed compounds in the model space. Outliers were identified as being actively transported. The descriptors chosen for inclusion in the model were PSA and AlogP98, based on consideration of the physical processes involved in membrane permeability and the interrelationships and redundancies between available descriptors. These descriptors are quite straightforward for a medicinal chemist to interpret, enhancing the utility of the model. Molecular weight, while often used in passive absorption models, was shown to be superfluous, as it is already a component of both PSA and AlogP98. Extensive validation of the model on hundreds of known orally delivered drugs, "drug-like" molecules, and Pharmacopeia, Inc. compounds, which had been assayed for Caco-2 cell permeability, demonstrated a good rate of successful predictions (74-92%, depending on the dataset and exact criterion used).

A statistical-based approach to assessing the fidelity of combinatorial libraries encoded with electrophoric molecular tags. Development and application of tag decode-assisted single bead LC/MS analysis.

A statistical sampling protocol is described to assess the fidelity of libraries encoded with molecular tags. The methodology, termed library QA, is based on the combined application of tag decode analysis and single bead LC/MS. The physical existence of library compounds eluted from beads is established by comparing the molecular weight predicted by tag decode with empirical measurement. The goal of sampling is to provide information on overall library fidelity and an indication of the performance of individual library synthons. The minimal sampling size n for library QA is l0 x the largest synthon set. Data are reported as proportion (p) +/- lower and upper boundary (lb-ub) computed at the 95% confidence level (alpha = 0.05). As a practical demonstration, library QA was performed on a 25,200-member library of statine amides (size = 40 x 63 x 10). Sampling was conducted three times at n approximately 630 beads per run for a total of 1902 beads. The overall proportions found for the three runs were consistent with one another: p = 84.4%, lb-ub = 81.5-87.2%; p = 83.1%, lb-ub = 80.2-85.95; and p = 84.5%, lb-ub = 81.8-87.3%, suggesting the true value of p is close to 84% compound confirmation. The performance pi of individual synthons was also computed. Corroboration of QA data with biological screening results obtained from assaying the library against cathepsin D and plasmepsin II is discussed.

Outlier detection in multivariate analytical chemical data.

The unreliability of multivariate outlier detection techniques such as Mahalanobis distance and hat matrix leverage has been known in the statistical community for well over a decade. However, only within the past few years has a serious effort been made to introduce robust methods for the detection of multivariate outliers into the chemical literature. Techniques such as the minimum volume ellipsoid (MVE), multivariate trimming (MVT), and M-estimators (e.g., PROP), and others similar to them, such as the minimum covariance determinant (MCD), rely upon algorithms that are difficult to program and may require significant processing times. While MCD and MVE have been shown to be statistically sound, we found MVT unreliable due to the method's use of the Mahalanobis distance measure in its initial step. We examined the performance of MCD and MVT on selected data sets and in simulations and compared the results with two methods of our own devising. Both the proposed resampling by the half-means method and the smallest half-volume method are simple to use, are conceptually clear, and provide results superior to MVT and the current best-performing technique, MCD. Either proposed method is recommended for the detection of multiple outliers in multivariate data.