Pseudonatural Products Occur Frequently in Biologically Relevant Compounds

A new methodology for classifying fragment combinations and characterizing pseudonatural products (PNPs) is described. The source code is based on open-source tools and is organized as a Python package. Tasks can be executed individually or within the context of scalable, robust workflows. First, structures are standardized and duplicate entries are filtered out. Then, molecules are probed for the presence of predefined fragments. For molecules with more than one match, fragment combinations are classified. The algorithm considers the pairwise relative position of fragments within the molecule (fused atoms, linkers, intermediary rings), resulting in 18 different possible fragment combination categories. Finally, all combinations for a given molecule are assembled into a fragment combination graph, with fragments as nodes and combination types as edges. This workflow was applied to characterize PNPs in the ChEMBL database via comparison of fragment combination graphs with natural product (NP) references, represented by the Dictionary of Natural Products. The Murcko fragments extracted from 2000 structures previously described were used to define NP fragments. The results indicate that ca. 23% of the biologically relevant compounds listed in ChEMBL comply to the PNP definition and that, therefore, PNPs occur frequently among known biologically relevant small molecules. The majority (>95%) of PNPs contain two to four fragments, mainly (>95%) distributed in five different combination types. These findings may provide guidance for the design of new PNPs.

IV. Distribution shape analysis of the fragment molecule coverage S13 V. Impact of the benzene fragment on the results S13 VI. Computational time S15  Figure S7.  The npfc package is available at https://github.com/mpimp-comas here additional documentation is available on the package itself (API and overall architecture). The "_fcp_labels" property in the provided SDF is pickled and encoded as base64 strings. Please refer to the npfc.utils.decode_object to convert them back into Python dictionaries.

II. Manual curation of structural errors in the NP fragments
Five structures could not be parsed with RDKit directly using the input SDF. Rather than just discarding them, the structures were fixed using MarvinSketch 19.2.0 ( Figure S1).

III. Preparation of the datasets
The preparation of the datasets consisted of 5 steps: 1. Split the data into chunks (optional) 2. Load the SD File(s) into Pandas DataFrame(s) with RDKit molecules 3. Standardize the structures 4. Removing duplicate entries 5. Depict molecules During these steps, molecules might fail the process and be discarded (Table S1).

S5
Similarly, molecules could be filtered out if they did not respect certain criteria (Table S2). Step Label Description For fragments, only the filters on empty and duplicate structures were enabled, as well as the timeout limit (mandatory).

NP-derived fragments
The 2,000 fragments from Over et al. were prepared after five of them were manually curated (see section II). Results are represented by Figure S2 (overall) and Figure S3 (filtered entries).

Natural Products (DNP)
The results of the preparation of the 318,271 records of the DNP dataset are represented by Figure  S4 (overall), Figure S5 (filtered entries) and Figure S6 (errors).

Synthetic Compounds (ChEMBL)
The preparation of the 1,941,411 records of the ChEMBL26 dataset are represented by Figure S7 (overall), Figure S8 (filtered entries) and Figure S9 (errors).

IV. Distribution shape analysis of the fragment molecule coverage
This section refers to the figures 6e and 6f of the manuscript. Skewness and Kurtosis are already described in the main text and the corresponding code and results are located in the included "distribution_shape_stats.pdf" SI file. Since the distributions for NPs and PNPs when considering side chains (SC+) were bell-shaped, probability plots were computed to check it they could be Gaussian ( Figure S10). For comparison, results are presented for both datasets when side chains are not considered as well (SC-). In all cases, the data does not follow a Gaussian distribution.

V. Impact of the benzene fragment on the results
The data below describes the results obtained during our first attempt of the NPFC approach, while including the benzene as a Natural Product (NP) fragment. The datasets, software versions and computational resources were the same as used for the main manuscript. During our first attempt at the NPFC project, we included all fragments obtained from the Over et al. dataset. Since the Murcko scaffolds were extracted from the structures (using RDKit), the benzene ring was obtained as an NP-derived fragment on its own and was found to strongly impact the results (Table S3). Indeed, the benzene fragment was found to be part of 37.34% of all remaining NPs at the end of the workflow and represented 26.73% of all fragment combinations at this stage (data not shown). The impact observed for PNPs was even stronger, with the benzene ring found in 63.39% of the structures and representing 44.11% of all fragment combinations.