The chemfp project

Abstract

The chemfp project has had four main goals: (1) promote the FPS format as a text-based exchange format for dense binary cheminformatics fingerprints, (2) develop a high-performance implementation of the BitBound algorithm that could be used as an effective baseline to benchmark new similarity search implementations, (3) experiment with funding a pure open source software project through commercial sales, and (4) publish the results and lessons learned as a guide for future implementors. The FPS format has had only minor success, though it did influence development of the FPB binary format, which is faster to load but more complex. Both are summarized. The chemfp benchmark and the no-cost/open source version of chemfp are proposed as a reference baseline to evaluate the effectiveness of other similarity search tools. They are used to evaluate the faster commercial version of chemfp, which can test 130 million 1024-bit fingerprint Tanimotos per second on a single core of a standard x86-64 server machine. When combined with the BitBound algorithm, a k = 1000 nearest-neighbor search of the 1.8 million 2048-bit Morgan fingerprints of ChEMBL 24 averages 27 ms/query. The same search of 970 million PubChem fingerprints averages 220 ms/query, making chemfp one of the fastest CPU-based similarity search implementations. Modern CPUs are fast enough that memory bandwidth and latency are now important factors. Single-threaded search uses most of the available memory bandwidth. Sorting the fingerprints by popcount improves memory coherency, which when combined with 4 OpenMP threads makes it possible to construct an N × N similarity matrix for 1 million fingerprints in about 30 min. These observations may affect the interpretation of previous publications which assumed that search was strongly CPU bound. The chemfp project funding came from selling a purely open-source software product. Several product business models were tried, but none proved sustainable. Some of the experiences are discussed, in order to contribute to the ongoing conversation on the role of open source software in cheminformatics.

Keywords: FOSS; Format; High-performance; Molecular fingerprints; Open source; Performance benchmark; Similarity searching; Tanimoto.

Fig. 1

Example FPS file for 166-bit MACCS keys generated by OpenEye’s GraphSim toolkit. Header lines start with a ‘#’. The three record lines start with a hex-encoded fingerprint, followed by a tab and the record id

Fig. 2

Seven fingerprint type strings from different toolkits. Each type string contains space separated terms. The first term contains the fingerprint family name and version. Remaining terms encode fingerprint parameters as key = value pairs. The OpenEye-Path and RDKit-Morgan types are wrapped over two lines for presentation

Fig. 3

Single query search times for chemfp 3.3. Boxen plots for k = 2, 10, 100, and 1000 nearest-neighbor and threshold = 0.95, 0.80, 0.70, and 0.40 searches of ChEMBL 24 and PubChem (downloaded 2018-12-07). Each search samples 1000 fingerprints to use as queries so each query is always found in the result. Python’s garbage collector was disabled for each timing as it adds a roughly 25 ms delay about every 1000 timings. The T = 0.40 PubChem search could not be run due to insufficient memory

Fig. 4

Example of how the non-distributive nature of IEEE 754 doubles results in different Tversky similarity scores