Rendering Conventional Molecular Fingerprints for Virtual Screening Independent of Molecular Complexity and Size Effects

Abstract

Molecular complexity and size effects represent a known complication of fingerprint similarity searching and virtual screening that often leads to an increase in false-positive rates and a decrease in hit rates. In standard fingerprints, differences in the complexity of reference and database molecules lead to different fingerprint bit densities, which negatively affects similarity search calculations, in particular, when fingerprints of reference molecules have higher bit density than corresponding fingerprints of database compounds. In pharmaceutical research, this is the case in many practical virtual screening applications when chemically optimized reference molecules are used. Herein we introduce an intuitive computational method to make standard fingerprints such as structural keys or pharmacophore feature fingerprints independent of molecular complexity and size effects. This is achieved by applying the concept of 'balanced codes' originating in computer science. Following this approach, binary fingerprints are transformed by incorporating the complement of their bit patterns. This straightforward transformation produces fingerprint representations with characteristic bit patterns that have exactly half of their bit positions set on, corresponding to a constant bit density of 50 % for all test compounds, regardless of their complexity and size. In similarity search calculations in the presence of complexity effects of increasing magnitude, transformed structural key and pharmacophore fingerprints display consistently better performance than their unmodified counterparts and recover active compounds in cases where the original fingerprints fail.

Molecular Similarity in Medicinal Chemistry

“…36 Molecular complexity effects can be balanced or eliminated in different ways, for example, by equally taking into account bits that are set on or off in similarity calculations 37,38 or by combining binary fingerprint representations with their complements, i.e., adding the complement to the original bit string, thereby producing a constant fingerprint bit density for compounds of any size. 39 Calculating Tanimoto, Tversky, or Dice similarity has an assumed advantage that numerical values can now be used to distinguish similarity relationships in a consistent manner. How does this numerical approach from chemical informatics relate to, and perhaps influence, the more subjective assessment of similarity in medicinal chemistry?…”

FragNet, a Contrastive Learning-Based Transformer Model for Clustering, Interpreting, Visualizing, and Navigating Chemical Space

“…Its particular importance commonly comes in circumstances where one has a "hit" in a bioassay and wishes to select from a library of available molecules of known structure which ones to prioritize for further assays that might detect a more potent hit. The usual means of assessing molecular similarity are based on encoding the molecules as vectors of numbers based either on a list of measured or calculated biophysical or structural properties, or via the use of so-called molecular fingerprinting methods (e.g., [135][136][137][138][139][140][141][142]). We ourselves have used a variety of these methods in comparing the "similarity" between marketed drugs, endogenous metabolites and vitamins, natural products, and certain fluorophores [91,99,113,[143][144][145][146][147][148].…”

Advances in 2D fingerprint similarity searching

“…Thus, similarity search calculations with PDR-FP are not biased by differences in molecular complexity [78] and have been shown to be particularly effective on compound classes having high structural diversity where other types of fingerprints produce only low compound recall [32,80]. Furthermore, it has recently also been demonstrated that conventional keyed fingerprints can be rendered complexity-independent through 'balanced codes' transformation, that is, by merging a fingerprint with the complement of its bit setting, which generates a constant bit density of 50% (but doubles the size of the fingerprint) [81]. Fligner et al [82] introduced a modified version of the Tc that was able to reduce the compound size bias in the selection of diverse subsets from source libraries.…”