Differential Multimolecule Fingerprint for Similarity Search─Making Use of Active and Inactive Compound Sets in...

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Traditional molecular similarity calculations rely on numerical criteria like the Tanimoto index.
• Virtual screening requires query molecules to be highly representative of the target compound class.

Purpose of the Study:

• To introduce a novel differential multimolecule fingerprint (DMMFP) for more specific and sensitive virtual screening.
• To evaluate the performance of DMMFP against established methods and diverse datasets.

Main Methods:

• Developed a DMMFP by weighting features based on frequency in active and inactive molecules, then subtracting inactive from active values.
• Employed a modified Sørensen-Dice coefficient for similarity computation with the DMMFP.
• Tested the DMMFP approach on five diverse compound classes with known binding affinities, including decoy and dark chemical matter datasets.

Main Results:

• DMMFP demonstrated high sensitivity in recovering active molecules using MACCS and PubChem fingerprints.
• Successfully identified known drugs and inhibitors within the dark chemical matter dataset.
• The DMMFP combined with a Bayesian classifier showed superior specificity and accuracy; the Sørensen-Dice coefficient highlighted potential for early active molecule recognition.

Conclusions:

• The DMMFP approach offers a significant advancement in virtual screening by generating class-specific fingerprints.
• It effectively identifies active compounds, even within challenging datasets like dark chemical matter.
• The method provides a valuable tool for drug discovery, enhancing both sensitivity and specificity.