Visualizing high dimensional descriptors of molecular structures
Abstract
The distribution of chemical compounds in high-dimensional molecular descriptor space can be viewed in two dimensions by applying the projection method of this invention. This method has particular usefulness for viewing the relationships of a large number of compounds such as found in a large scale HTS or virtual combinatorial library. After selecting a representative subset of the larger data set of compounds, initially components from the high-dimensional descriptor space are determined by PCA. In order to relax an NLM projection using the PCA components as a start, the stress function is modified to reflect a local horizon beyound which the separation of the compounds is not meaningfully measureable. The resulting two dimensional projections provide a clear insight into the distribution of the chemical compounds in the higher dimensional space. The method is clearly generalizable to viewing descriptor space in three dimensions and to using high dimensional descriptors other than those used to describe molecular structure.
Claims
exact text as granted — not AI-modifiedWhat is claimed is:
1 . A method of visualizing in two or three dimensions the relationships in diversity space of compounds which are characterized by high-dimensional descriptors comprising the steps of:
a. selecting a representative subset of the compounds; b. calculating or retrieving from a data base the descriptor characteristics for each compound of the subset selected; c. generating a distance matrix between all compounds or utilizing a function to generate the distance matrix elements as needed; d. computing a hierarchy of clusters, defining cluster centers and partitioning each set of compounds at each level of clustering; e. performing a PCA projection onto the first two components; f. using a modified stress function which reflects a horizon, running an NLM refinement from the initial PCA coordinates; and g. graphically displaying the coordinates of each compound as determined by the NLM refinement.Join the waitlist — get patent alerts
Track US2004241709A1 — get alerts on status changes and closely related new filings.
We store only your email — no account needed. See our privacy policy.