Skip to main content
Download PDF
Download ePub

One chiral fingerprint to find them all

Journal of Cheminformatics volume 16, Article number: 53 (2024) Cite this article

Abstract

Molecular fingerprints are indispensable tools in cheminformatics. However, stereochemistry is generally not considered, which is problematic for large molecules which are almost all chiral.

Herein we report MAP4C, a chiral version of our previously reported fingerprint MAP4, which lists MinHashes computed from character strings containing the SMILES of all pairs of circular substructures up to a diameter of four bonds and the shortest topological distance between their central atoms. MAP4C includes the Cahn-Ingold-Prelog (CIP) annotation (R, S, r or s) whenever the chiral atom is the center of a circular substructure, a question mark for undefined stereocenters, and double bond cis–trans information if specified. MAP4C performs slightly better than the achiral MAP4, ECFP and AP fingerprints in non-stereoselective virtual screening benchmarks. Furthermore, MAP4C distinguishes between stereoisomers in chiral molecules from small molecule drugs to large natural products and peptides comprising thousands of diastereomers, with a degree of distinction smaller than between structural isomers and proportional to the number of chirality changes. Due to its excellent performance across diverse molecular classes and its ability to handle stereochemistry, MAP4C is recommended as a generally applicable chiral molecular fingerprint.

Scientific contribution

The ability of our chiral fingerprint MAP4C to handle stereoisomers from small molecules to large natural products and peptides is unprecedented and opens the way for cheminformatics to include stereochemistry as an important molecular parameter across all fields of molecular design.

Graphical Abstract

Introduction

Many computational tasks related to small molecule drug discovery, such as similarity searches [1, 2], target prediction [3,4,5,6,7], ligand-based virtual screening [8] and visualization of large databases of drug-like molecules [9,10,11,12,13,14,15,16,17,18], can be performed using vectors encoding molecular structure, called molecular fingerprints [19, 20]. Remarkably, molecular fingerprints work quite well to classify and compare bioactive molecules without considering stereochemical information, which is somewhat surprising considering that biological matter is essentially chiral and stereo-defined at the molecular level [21,22,23], but also reflects the fact one only rarely needs to distinguish between different stereoisomers of small molecule drugs, in part simply because many drug-like compounds are achiral.

In the context of developing computational tools for new modalities including beyond-Ro5 molecules [24, 25], in our case for peptides with variable chain topology and stereochemistry [26,27,28], we have adapted molecular fingerprints based on atom-pairs [29,30,31,32] for large molecules such as peptides and proteins [33,34,35]. In particular, we combined atom-pair analysis and circular substructures as encoded by the Morgan fingerprint ECFP4 [36, 37], with the principle of data compression using MinHashing [38,39,40,41], to design MAP4, a MinHashed Atom-Pair fingerprint. MAP4 encodes all possible pairs of circular substructures up to a diameter of four bonds in a molecule [42]. These pairs are written in the form of two canonicalized SMILES [43, 44] separated by the shortest topological distance, counted in bonds, between the corresponding pair of central atoms. Remarkably, MAP4 distinguishes molecular structures across different compound classes spanning from small molecules to natural products, peptides and the metabolome, for which other fingerprints such as the classical Morgan (ECFP4) [37] and Atom Pair (AP) [29] fingerprints fall short. In addition, MAP4 outperforms these and many other fingerprints in virtual screening benchmarks for both small molecule drugs [20] and peptides [42].

Similarly to commonly used molecular fingerprints however, MAP4 does not include stereochemistry (cis–trans double bonds, enantiomers and diastereomers), which is clearly an omission considering that most molecules beyond Ro5, such as diverse natural products and synthetic compounds in the public databases ChEMBL [45], COCONUT [46], and ZINC [47], are chiral (Fig. 1a). To correct this omission and enable the cheminformatic analysis of compounds with multiple chiral centers such as carbohydrates and peptides, we now report MAP4C, an improved version of the MAP4 fingerprint. MAP4C includes the description of chiral centers following the Cahn-Ingold-Prelog (CIP) nomenclature in a fraction of molecular shingles (Fig. 1b, c), as well as double bond stereochemistry.

Fig. 1
figure 1

Molecular chirality and fingerprints. a Correlation between chirality and heavy atom count (HAC) across ChEMBL, COCONUT, and ZINC datasets. The blue line depicts the percentage of chiral molecules relative to HAC. A steady increase in the percentage of chiral molecules is observed with increasing HAC. The yellow line represents the total count of molecules corresponding to each HAC. b Chiral shingle generation concept exemplified on a selected atom pair of polymyxin B2. The generated shingle corresponds to the pair of circular substructures (blue) separated by the shortest topological distance (red) of their central atoms. Whenever the central atom of a substructure is chiral, the atom symbol in the substructure SMILES is replaced by the Cahn-Ingold-Prelog (CIP) descriptor (R, S, r, or s), or by a question mark (?) if the stereochemistry is not defined, bracketed by two “$” characters (yellow). c Percentage of molecular shingles containing chiral information vs. percentage of chiral atoms in the molecule for MAP4C (largest diameter of four bonds). These percentages were computed using a dataset of chiral molecules uniformly sampled from the Riniker & Landrum benchmark. The high r2 and Pearson correlation coefficients underscore a strong association between the two variables

Full size image

Methods

Fingerprint design

The chiral version of the MinHashed Atom-Pair fingerprint (MAPC) was implemented in Python using RDKit following these steps:

  1. 1.

    At every non-hydrogen atom, extract all circular substructures up to the specified maximum radius as isomeric, canonical SMILES. Isomeric information (“@” and “@@” characters) is manually removed from the extracted SMILES, while the implicit E/Z-isomerism (“/”, and “\” characters) are maintained. Allene chirality and conformational chirality such as in biaryls or in helicenes are not considered, as they cannot be specified in the SMILES notation. Radius 0 is skipped.

  2. 2.

    At the specified maximum radius, whenever the central atom of a circular substructure is chiral, replace the first atom symbol in the extracted SMILES with its Cahn-Ingold-Prelog (CIP) descriptor bracketed by two “$” characters ($CIP$). The CIP descriptor of the chiral atom is defined on the entire molecule, not on the extracted substructure.

  3. 3.

    At each radius, generate shingles for all possible pairs of extracted substructures. Each shingle contains two substructures and their topological distance in following format: “substructure 1 | topological distance | substructure 2”.

  4. 4.

    MinHash the list of shingles to obtain a fixed sized vector. The MinHashing procedure is explained in detail in our previous publication [38, 42].

Benchmark

The virtual screening performance of the MAPC fingerprint was evaluated in a comparative study with commonly used fingerprints (ECFP4 [37], ECFP6 [37], Atom-Pair [29]) in a benchmark adapted from Riniker and Landrum [20]. Since the structure SMILES in the original benchmark do not contain any stereochemistry, the respective chiral SMILES (when applicable) were retrieved from the DUD [48], MUV [49] and ChEMBL [45] databases using the provided compound IDs.

Additional 60 peptide sets were included in the benchmark to test the performances of the fingerprints for large biomolecules. For each of 30 random linear sequences, a set containing 10,000 single-point mutants and a set containing 10,000 scrambled versions of the random sequence were generated and BLAST analogues labelled as actives. The precise generation procedure of the peptide datasets is described in our previous publication [42].

For every set, 5 randomly selected actives were extracted and stored in a separate file. The mean and standard deviation of pairwise ECFP4C Tanimoto and MAP4C Jaccard similarities of the five selected actives are reported in the Additional file 1: Figures S1, S2. Each of the selected actives was used as a query to rank the remaining compounds in the set based on fingerprint similarity (Jaccard similarity for MinHashed fingerprints; Dice similarity for folded fingerprints). AUC, EF1, EF5, BEDROC20, BEDROC100, RIE20 and RIE100 metrics were calculated for the obtained ranked lists and averaged along the 5 queries for every set in the benchmark. Additionally, the fingerprints were ranked based on the obtained performance metrics and finally the average rank of each fingerprint determined for all metrics. Pearson correlation coefficients and Friedman-Nemenyi post-hoc tests were calculated for all fingerprint pairs using the scipy and scikit-posthocs Python libraries.

Stereoisomers, isomers and scrambled sequences

We enumerated all possible stereoisomers of molecules 1–14 (Figs. 1c and  4) by generating all possible isomeric SMILES combinations, canonicalizing them, and removing duplicates. We additionally enumerated all possible permutations of ln65 (7) and polymyxin B2 (1) sequences, obtaining a total of 330 and 1,512 scrambled sequences respectively. Structural isomers of 1,4-diaminocyclohexane (15) and aminopiperazine (16) were extracted from GDB-13 using the MQN-browser [50, 51]. The extracted sets contained 203 structural isomers of 15, of which 156 contained one or more stereocenters and 48 structural isomers of 16, of which 29 contained one or more stereocenters. For each structural isomer, all possible stereoisomers were generated using the RDKit “EnumerateStereoisomers” function, yelding 746 unique structures for 15 and 126 for 16. For all stereoisomers and permutations, fingerprints were calculated as 2048-bit vectors.

TMAP

The indices obtained from the MAP4C calculation were used to create a locality-sensitive hashing (LSH) forest of 32 trees. For each molecular structure, the 500 approximate nearest neighbors in the MAP4C feature space were extracted from the LSH forest and used to calculate the TMAP layout [16]. The resulting layout was displayed in an interactive TMAP using the open-source Faerun package [15].

Results and discussion

Encoding stereochemistry in MAP fingerprints

The MAP (MinHashed Atom-Pair) fingerprint of a molecule consists in a series of MinHashes computed from the list of its molecular shingles [38,39,40,41]. A molecular shingle is written for each possible pair of circular substructures of a given diameter (2 bonds for MAP2, 4 bonds for MAP4, 6 bonds for MAP6), written as canonicalized SMILES, separated by the shortest topological distance separating the central atoms, counted in bonds [42]. We preserve the Z/E double bond information in all shingles whenever the entire double bond is included in a shingle. To encode stereocenter information into our fingerprints, we label chiral atoms with their Cahn–Ingold–Prelog (CIP) descriptor (R, S, r or s), as computed by RDKit, whenever stereochemistry is defined, or label them with a question mark (“?”) if stereochemistry is not specified. Importantly, we only apply the chiral label when a chiral atom is the central atom of a circular substructure and only for shingles with the largest diameter considered. The concept is illustrated for one of the possible pairs involving the stereocenter in polymyxin B2 (1, Fig. 1b).

When applied to a dataset of chiral molecules uniformly sampled from the Riniker and Landrum benchmark (Additional file 1: Figure S3) [20], we find that the percentage of molecular shingles containing chiral information is approximately the same as the percentage of chiral atoms in a molecule for MAP2C (largest diameter of two bonds, Additional file 1: Figure S4a), MAP4C (largest diameter of four bonds, Fig. 1c) and MAP6C (largest diameter of six bonds, Additional file 1: Figure S4b). Most importantly, chiral information only appears in a relatively small fraction of all possible shingles, such that any defined stereoisomer of a molecule has a relatively high similarity to the molecule without assigned stereochemistry, for which the MAPC fingerprint is identical to the MAP fingerprint.

Virtual screening benchmark

The relevance of any molecular fingerprint for drug discovery can be tested by attempting to retrieve known bioactive compounds for a given target by nearest-neighbor searches from one of the known active compounds in a dataset in which the known actives have been mixed with so-called decoys. These decoys are molecules selected randomly from databases to have similar physico-chemical properties as the actives, but which are not documented to be active on the target. Here we used the reference benchmarking dataset of Riniker and Landrum for small molecule drugs [20], which considers 118 active and decoy datasets taken from DUD [48], MUV [49], and ChEMBL [45]. For larger molecules, we used our previously reported set of 60 different randomly chosen 10−, 15− and 20−mer peptides mixed with either random single point mutants (30 sets), or sequence scrambled analog (30 sets) [42], for which we challenge the fingerprint to retrieve BLAST search analogs [52].

We compared the performance of MAP2C, MAP4C, and MAP6C with their respective achiral counterparts, as well as with reference binary fingerprints ECFP4, ECFP6, and AP, and their corresponding chiral versions (ECFP4C, ECFP6C, and APC). The primary objective of the benchmark experiment was to ensure that the inclusion of chirality does not compromise the baseline virtual screening capabilities of the original MAP fingerprint. Indeed, fingerprints in their chiral and non-chiral versions demonstrated comparable performances across various test sets and performance metrics, showing that including chirality information was not detrimental to fingerprint performance in these benchmarks (Fig. 2a, b and Additional file 1: Figure S5–S9).

Fig. 2
figure 2

Virtual Screening benchmark (a) AUC and (b) EF1 of MAP6 (purple), MAP4 (magenta), MAP2 (blue), AP (grey), ECFP6 (orange) and ECFP4 (yellow) and across all small molecules and peptide targets (80 ChEMBL targets, 21 DUD targets, 17 MUV targets, 30 mutated peptide targets, and 30 scrambled peptide targets). Chiral fingerprints are displayed as bold lines, non-chiral fingerprints are displayed as dashed lines. The value displayed for each dataset is the mean metric of 5 runs. c Mean ranks of fingerprints across all virtual screening datasets for each metric. Small molecule sets (ChEMBL, DUD, MUV) and peptide sets are presented separately to highlight the differences in relative performance

Full size image

The ranks of the different fingerprints for the various performances measures showed that the MAP fingerprints performances were slightly ahead of the other fingerprints, with MAP4C appearing with the best ranks in the small molecule benchmark and MAP6C in the peptide benchmark (Fig. 2c). However, a pairwise Friedman-Nemenyi test across all performance metrics showed that the difference between chiral over non-chiral fingerprints of each type (MAPC vs. MAP, ECFPC vs. ECFP and APC vs AP) was not significant (Additional file 1: Figures S10-16). The only statistically significant differences were between groups. For instance, MAP(C) fingerprints significantly outperformed ECFP(C) and AP(C) fingerprints with exception of AP(C) for the AUC metric. MAP(C) fingerprints combine high local precision of circular substructure encoding, akin to ECFPs, with the perception of atom pairs reflecting global structural features, akin to AP fingerprints. This combination is particularly effective in scenarios where both local precision and global structure are relevant to differentiate between active and non-active molecules, possibly explaining the higher performance of the MAP(C) fingerprints compared to ECFP(C) and AP(C).

Finding all stereoisomers

In addition to be on par with non-chiral fingerprints for the above virtual screening benchmarks, one would expect a chiral fingerprint to distinguish all possible stereoisomers of a chiral molecule. To test the chiral differentiation of our fingerprints, we investigated their ability to assign a different fingerprint value for each stereoisomer on a series of stereochemically complex molecules comprizing carbohydrates, peptides and macrocyclic natural products containing up to thousands of stereoisomers per molecule (Fig. 3 and Table 1).

Fig. 3
figure 3

Structures of natural products and peptides selected for the stereoisomer distinction task

Full size image
Table 1 Stereoisomer and scrambled sequence distinction task for selected natural products and peptides with multiple chiral centers and varying degrees of internal symmetry
Full size table
Fig. 4
figure 4

Differentiation between stereoisomers and structural isomers, shown as box plots of average Jaccard distances between pairs of stereoisomers (blue) or structural/sequence isomers (yellow). a structural isomers of 1,4-diaminocyclohexane (203) and 4-aminopiperidine (48) and their diastereomers. The skewed distribution of Jaccard distance of 15 with MAP6C is caused by two outliers exhibiting a distance of 0 which cannot be represented on the log scale and is likely due to a bit-clash issue. b sequence isomers (330) or diastereomers (2,048) of ln65 (7) as function of the Levenshtein distance separating each pair. c sequence isomers (1,512) or diastereomers (512) of polymyxin B2 (1) as function of the Levensthein distance separating each pair. See Figures S10 and S11 for plots with MAP6C, MAP2C and ECFP6C. See methods for details

Full size image

For carbohydrates, both MAP6C and MAP4C readily distinguished the 32 stereoisomers of α-D-glucopyranose (2), the 1024 stereoisomers of the disaccharide lactose (3), the 528 possible stereoisomers of the non-reducing, C2-symmetrical α-diglucoside trehalose (4), the 16,384 stereoisomers of the aminoglycoside antibiotic validamycin A (5), and the nine possible stereoisomers of the signaling carbocyclic sugar myo-inositol (6). By contrast, the four other chiral fingerprints tested all fell short in at least one of the six cases, and APC failed on all of them.

Our MinHashed fingerprints performed very well with peptide stereoisomers. In the case of the antimicrobial undecapeptide ln65 (7), a membrane disruptive antimicrobial peptide whose activity/toxicity balance is modulated by stereochemical variations and which motivated the present study [28], the three chiral MAP fingerprints distinguished all the 2,048 possible stereoisomers. By contrast, ECFP6C only saw about half of them and ECFP4C and APC distinguished less than 10%, most likely because this peptide is composed of only lysine and leucine residues, which reduces the number of possible substructures. The chiral MAP fingerprints also distinguished the 330 possible sequence-scrambled isomers of 7 and the 675,840 possible stereoisomers of sequence-scrambled isomers of 7. By comparison, APC succeeded for the 330 scrambled sequences but failed on the larger set, and both chiral ECFPs failed in both cases, which can be attributed to the absence of long-range substructures in ECFP fingerprints.

The ability of chiral MAP fingerprints to perceive peptide stereoisomers was also well illustrated by their ability to distinguish all 512 stereoisomers of the cell-penetrating peptide nona-arginine (8) [53, 54], as well as the 4096 stereoisomers of polymyxin B2 (1), used as last resort antibiotic against multidrug resistant bacteria [55]. In the latter case, our fingerprints also distinguished between the 1,512 possible sequence-scrambled isomers of 1, the 774,144 possible sequence-scrambled stereoisomers of 1, as well as between the 531,441 possible assignments of chirality as R, S, or undefined stereochemistry in the 12 chiral centers of 1. An undefined stereochemistry corresponds to a stereorandomized position accessible by chemical synthesis using a racemic amino acid at that position (stereorandomization at multiple position can lead to partially active analogs as reported for 1) [56]. In all of these cases, APC and ECFPCs were unable to distinguish all possibilities.

Macrocyclic natural products with rotational symmetries were particularly challenging for chiral fingerprints. For instance, only MAP4C and MAP6C correctly identified the 136 possible stereoisomers of the cyclic peptide antibiotic quinaldopeptin (9) and the 2,080 stereoisomers of the cytotoxic macrocyclic depsipeptide onchidin (10), two natural product macrocycles with C2 symmetry. By contrast, the 528 stereoisomers of the C2 symmetrical antimicrobial macrocyclic peptide gramicidin S (11) were only distinguished by MAP6C. Furthermore, none of the chiral fingerprints tested was able to cope with the C3 symmetrical dodecadepsipeptide antibiotic valinomycin (12, 1,376 stereoisomers), the C4 symmetrical macrolide ionophore antibiotic nonactin (13, 16,456 stereoisomers), or the C7 symmetrical hepta-arginine cyclic peptide NP213 developed as antifungal agent (14, 20 stereoisomers). Note that all fingerprints were used with 2,048-bits, but that performance did not increase significantly when using much larger bit sizes or without MinHashing or folding.

Ranking stereoisomers versus isomers

The degree of differentiation between stereoisomers should be proportional to the number of stereochemical changes between any two stereoisomers, and should also be smaller than the difference to a different molecule such as a structural isomer. We tested the ability of our chiral fingerprints for this task for small and large molecules separately. As a test case for small molecules, we computed Jaccard distances between all pairs involving the 203 structural isomers of 1,4-diaminocyclohexane (15), a ring fragment which is enriched in bioactive molecules from ChEMBL [57, 58], and between all pairs of stereoisomers in the set. We similarly analyzed all pairs involving the 48 structural isomers of 4-aminopiperazine (16), a similar drug scaffold, and the stereoisomeric pairs within the set. Generally, MAPC distances were higher than those of other fingerprints. This outcome is unsurprising, given that MAPC encodes a notably greater number of features, which also contributes to its high precision. In both test cases, all six fingerprints ranked pairs stereoisomers closer to each other than pairs of structural isomers (Fig. 4a/b).

For peptides, we measured Jaccard distances between pairs of scrambled-sequence isomers versus pairs of stereoisomers with the same sequence for ln65 (7) and polymyxin B2 (1). For peptides, the degree of sequence similarity can also be measured by the Levenshtein distance, which represents the minimum number of mutations necessary to transform one sequence into another one, considering residue type changes, stereochemical inversions, insertions and deletions (Fig. 4c/d and Additional file 1: Figure S17, S18). Jaccard distances generally increased with increasing Levensthein distances for all fingerprints. Similar to small molecules, distances between peptide stereoisomers were smaller than between sequence isomers only for chiral MAP fingerprints and APC. However, chiral ECFPs assigned larger distances to stereoisomers than to sequence isomers, which probably relates to their inability to distinguish many pairs of sequence isomers. For both ln65 (7) and polymyxin B2 (1), the lower Jaccard distances between stereoisomers compared to sequence isomers was well visible in TMAP representations of each dataset constructed using MAP4C as similarity measure (Fig. 5a/b) [16]. In both cases, there was a complete separation between the 2,048/512 stereoisomers of the parent peptide and the 330/1,512 sequence isomers.

Fig. 5
figure 5

MAP4C TMAPs showing the Jaccard distance (dJ; rainbow) of stereoisomers (blue) and sequence isomers (yellow) towards their respective queries: (a) ln65, 2,048 diastereomers and 330 sequence isomers. The interactive version of the TMAP is accessible under https://tm.gdb.tools/map4/MAP4C_ln65/ (b) polymyxin B2, 512 diastereomers and 1,512 sequence isomers. The interactive version of the TMAP is accessible under https://tm.gdb.tools/map4/MAP4C_pmb2/

Full size image

Conclusions

In summary, the data above shows that the chiral versions of MAP fingerprints reported here perform as good as their achiral versions in non-stereoselective virtual screening benchmarks. Remarkably, our chiral MAP fingerprints are able to distinguish stereoisomers even in cases involving up to thousands of stereoisomers where the chiral versions of ECFP and AP do not perform well. Furthermore, the chiral MAP Jaccard distances between enantiomers or stereoisomers are generally shorter than for structural isomers, allowing to use chiral MAP fingerprints as a refinement of their achiral version. Because MAP4C computes faster than MAP6C due to the small number of atom pairs considered, we recommend MAP4C as the molecular fingerprint of choice for comparing molecules spanning from small drug-like building blocks to large natural products and peptides. The ability of our chiral fingerprint MAP4C to handle stereoisomers from small molecules to large natural products and peptides is unprecedented and opens the way for cheminformatics to include stereochemistry as an important molecular parameter across all fields of molecular design.

Availability of data and materials

The source codes and datasets used for this study are available at https://zenodo.org/records/10389905 The code for MAPC can be found at https://github.com/reymond-group/mapchiral.

Abbreviations

AP(C):

Atom pair fingerprint (chiral)

AUC:

Area under the curve

BEDROC:

Boltzmann-enhanced discrimination of the receiver operating characteristic

BLAST:

Basic local alignment search tool

CIP:

Cahn-Ingold-Prelog

DUD:

Directory of useful decoys

ECFP(C):

Extended connectivity fingerprint (chiral)

EF:

Enrichment factor

FP:

Fingerprint

GDB:

Generated database

HAC:

Heavy atom count

JD:

Jaccard distance

LSH:

Locality sensitive hashing

MAP(C):

MinHash atom pair fingerprint (chiral)

MQN:

Molecular quantum numbers

MUV:

Maximum unbiased validation data sets

PMB2:

Polymyxin B2

RIE:

Robust initial enhancement

Ro5:

Rule of fives

ROC:

Receiver operating characteristic

SMILES:

Simplified molecular-input line-entry system

TMAP:

Tree map

References

  1. Willett P, Barnard JM, Downs GM (1998) Chemical similarity searching. J Chem Inf Comput Sci 38(6):983–996. https://doi.org/10.1021/ci9800211

    Article  CAS  Google Scholar 

  2. Maggiora G, Vogt M, Stumpfe D, Bajorath J (2014) Molecular similarity in medicinal chemistry. J Med Chem 57(8):3186–3204. https://doi.org/10.1021/jm401411z

    Article  CAS  PubMed  Google Scholar 

  3. Lagunin A, Stepanchikova A, Filimonov D, Poroikov V (2000) PASS: prediction of activity spectra for biologically active substances. Bioinformatics 16(8):747–748. https://doi.org/10.1093/bioinformatics/16.8.747

    Article  CAS  PubMed  Google Scholar 

  4. Keiser MJ, Roth BL, Armbruster BN, Ernsberger P, Irwin JJ, Shoichet BK (2007) Relating protein pharmacology by ligand chemistry. Nat Biotechnol 25(2):197–206. https://doi.org/10.1038/nbt1284

    Article  CAS  PubMed  Google Scholar 

  5. Czodrowski P, Bolick W-G (2016) OCEAN: optimized cross reactivity estimation. J Chem Inf Model 56(10):2013–2023. https://doi.org/10.1021/acs.jcim.6b00067

    Article  CAS  PubMed  Google Scholar 

  6. Mayr A, Klambauer G, Unterthiner T, Steijaert M, Wegner JK, Ceulemans H, Clevert D-A, Hochreiter S (2018) large-scale comparison of machine learning methods for drug target prediction on ChEMBL. Chem Sci 9(24):5441–5451. https://doi.org/10.1039/c8sc00148k

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Awale M, Reymond JL (2019) Web-based tools for polypharmacology prediction. Methods Mol Biol 1888:255–272. https://doi.org/10.1007/978-1-4939-8891-4_15

    Article  CAS  PubMed  Google Scholar 

  8. Scior T, Bender A, Tresadern G, Medina-Franco JL, Martinez-Mayorga K, Langer T, Cuanalo-Contreras K, Agrafiotis DK (2012) Recognizing pitfalls in virtual screening: a critical review. J Chem Inf Model 52(4):867–881. https://doi.org/10.1021/ci200528d

    Article  CAS  PubMed  Google Scholar 

  9. Schuffenhauer A, Ertl P, Roggo S, Wetzel S, Koch MA, Waldmann H (2007) The scaffold tree − visualization of the scaffold universe by hierarchical scaffold classification. J Chem Inf Model 47(1):47–58. https://doi.org/10.1021/ci600338x

    Article  CAS  PubMed  Google Scholar 

  10. Ertl, P.; Rohde, B. 2012 The Molecule Cloud - Compact Visualization of Large Collections of Molecules. J. Cheminf. 4 (1), Article 12. http://www.jcheminf.com/content/4/1/12 Accessed Dec 6, 2012.

  11. Lachance H, Wetzel S, Kumar K, Waldmann H (2012) Charting, navigating, and populating natural product chemical space for drug discovery. J Med Chem 55(13):5989–6001. https://doi.org/10.1021/jm300288g

    Article  CAS  PubMed  Google Scholar 

  12. Ruddigkeit L, Blum LC, Reymond JL (2013) Visualization and virtual screening of the chemical universe database GDB-17. J Chem Inf Model 53(1):56–65. https://doi.org/10.1021/ci300535x

    Article  CAS  PubMed  Google Scholar 

  13. Sander T, Freyss J, von Korff M, Rufener C (2015) DataWarrior: an open-source program for chemistry aware data visualization and analysis. J Chem Inf Model 55(2):460–473. https://doi.org/10.1021/ci500588j

    Article  CAS  PubMed  Google Scholar 

  14. Zhang B, Vogt M, Maggiora GM, Bajorath J (2015) Design of chemical space networks using a tanimoto similarity variant based upon maximum common substructures. J Comput-Aided Mol Des 29(10):937–950. https://doi.org/10.1007/s10822-015-9872-1

    Article  CAS  PubMed  Google Scholar 

  15. Probst D, Reymond J-L (2018) FUn: a framework for interactive visualizations of large. High Dimens Datasets Web Bioinformat 34(8):1433–1435. https://doi.org/10.1093/bioinformatics/btx760

    Article  CAS  Google Scholar 

  16. Probst D, Reymond J-L (2020) Visualization of very large high-dimensional data sets as minimum spanning trees. J Cheminform 12(1):12. https://doi.org/10.1186/s13321-020-0416-x

    Article  PubMed  PubMed Central  Google Scholar 

  17. Medina-Franco JL, Sánchez-Cruz N, López-López E, Díaz-Eufracio BI (2022) Progress on open chemoinformatic tools for expanding and exploring the chemical space. J Comput Aided Mol Des 36(5):341–354. https://doi.org/10.1007/s10822-021-00399-1

    Article  CAS  PubMed  Google Scholar 

  18. Zabolotna Y, Bonachera F, Horvath D, Lin A, Marcou G, Klimchuk O, Varnek A (2022) Chemspace atlas: multiscale chemography of ultralarge libraries for drug discovery. J Chem Inf Model 62(18):4537–4548. https://doi.org/10.1021/acs.jcim.2c00509

    Article  CAS  PubMed  Google Scholar 

  19. Willett P (2006) Similarity-based virtual screening using 2D fingerprints. Drug Discov Today 11(23–24):1046–1053. https://doi.org/10.1016/j.drudis.2006.10.005

    Article  CAS  PubMed  Google Scholar 

  20. Riniker S, Landrum GA (2013) Open-source platform to benchmark fingerprints for ligand-based virtual screening. J Cheminf 5(1):26. https://doi.org/10.1186/1758-2946-5-26

    Article  CAS  Google Scholar 

  21. Blackmond DG (2019) The origin of biological homochirality. Cold Spring Harb Perspect Biol 11(3):a032540. https://doi.org/10.1101/cshperspect.a032540

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Gal J (2013) Molecular chirality in chemistry and biology: historical milestones. Helv Chim Acta 96(9):1617–1657. https://doi.org/10.1002/hlca.201300300

    Article  CAS  Google Scholar 

  23. Benner SA (2017) Detecting darwinism from molecules in the enceladus plumes, jupiter’s moons, and other planetary water lagoons. Astrobiology 17(9):840–851. https://doi.org/10.1089/ast.2016.1611

    Article  PubMed  PubMed Central  Google Scholar 

  24. Waldmann H, Valeur E, Gueret SM, Adihou H, Gopalakrishnan R, Lemurell M, Grossmann TN, Plowright AT (2017) New modalities for challenging targets in drug discovery. Angew Chem Int Ed Engl 56:10294–10323. https://doi.org/10.1002/anie.201611914

    Article  CAS  PubMed  Google Scholar 

  25. Caron G, Digiesi V, Solaro S, Ermondi G (2020) Flexibility in early drug discovery: focus on the beyond-rule-of-5 chemical space. Drug Discov Today. https://doi.org/10.1016/j.drudis.2020.01.012

    Article  PubMed  Google Scholar 

  26. Di Bonaventura I, Jin X, Visini R, Probst D, Javor S, Gan BH, Michaud G, Natalello A, Doglia SM, Kohler T, van Delden C, Stocker A, Darbre T, Reymond JL (2017) Chemical space guided discovery of antimicrobial bridged bicyclic peptides against pseudomonas aeruginosa and its biofilms. Chem Sci 8(10):6784–6798. https://doi.org/10.1039/c7sc01314k

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Cai X, Orsi M, Capecchi A, Köhler T, Delden C, van Javor S, Reymond JL (2022) An intrinsically disordered antimicrobial peptide dendrimer from stereorandomized virtual screening. Cell Rep Phys Sci. https://doi.org/10.1016/j.xcrp.2022.101161

    Article  PubMed  PubMed Central  Google Scholar 

  28. Personne H, Paschoud T, Fulgencio S, Baeriswyl S, Köhler T, van Delden C, Stocker A, Javor S, Reymond J-L (2023) To fold or not to fold: diastereomeric optimization of an α-helical antimicrobial peptide. J Med Chem 66(11):7570–7583. https://doi.org/10.1021/acs.jmedchem.3c00460

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  29. Carhart RE, Smith DH, Venkataraghavan R (1985) Atom pairs as molecular features in structure-activity studies: definition and applications. J Chem Inf Comput Sci 25(2):64–73. https://doi.org/10.1021/ci00046a002

    Article  CAS  Google Scholar 

  30. Schneider G, Neidhart W, Giller T, Schmid G (1999) “Scaffold-hopping” by topological pharmacophore search: a contribution to virtual screening. Angew Chem Int Ed Engl 38(19):2894–2896

    Article  CAS  PubMed  Google Scholar 

  31. Awale M, Reymond JL (2014) Atom pair 2D-Fingerprints perceive 3D-molecular shape and pharmacophores for very fast virtual screening of ZINC and GDB-17. J Chem Inf Model 54:1892–1897. https://doi.org/10.1021/ci500232g

    Article  CAS  PubMed  Google Scholar 

  32. Awale M, Jin X, Reymond JL (2015) Stereoselective virtual screening of the zinc database using atom pair 3D-fingerprints. J Cheminf 7:3

    Article  Google Scholar 

  33. Jin X, Awale M, Zasso M, Kostro D, Patiny L, Reymond JL (2015) PDB-explorer: a web-based interactive map of the protein data bank in shape space. BMC Bioinformat 16:339. https://doi.org/10.1186/s12859-015-0776-9

    Article  CAS  Google Scholar 

  34. Capecchi A, Awale M, Probst D, Reymond JL (2019) pubchem and chembl beyond lipinski. Mol Inf 38:1900016. https://doi.org/10.1002/minf.201900016

    Article  CAS  Google Scholar 

  35. Orsi M, Probst D, Schwaller P, Reymond J-L (2023) Alchemical analysis of fda approved drugs. Digit Discov 2(5):1289–1296. https://doi.org/10.1039/D3DD00039G

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  36. Morgan HL (1965) The generation of a unique machine description for chemical structures-a technique developed at chemical abstracts service. J Chem Doc 5(2):107–113. https://doi.org/10.1021/c160017a018

    Article  CAS  Google Scholar 

  37. Rogers D, Hahn M (2010) Extended-connectivity fingerprints. J Chem Inf Model 50(5):742–754. https://doi.org/10.1021/ci100050t

    Article  CAS  PubMed  Google Scholar 

  38. Probst D, Reymond J-L (2018) A probabilistic molecular fingerprint for big data settings. J Cheminf 10(1):66. https://doi.org/10.1186/s13321-018-0321-8

    Article  CAS  Google Scholar 

  39. Broder AZ. 1998 On the Resemblance and Containment of Documents. In Proceedings. Compression and Complexity of SEQUENCES 1997 (Cat. No.97TB100171); IEEE Comput. Soc: Salerno, Italy. pp 21–29. https://doi.org/10.1109/SEQUEN.1997.666900.

  40. Manber U. 1994 Finding Similar Files in a Large File System. In Usenix Winter 1994 Technical Conference. pp 1–10.

  41. Damashek M (1995) Gauging similarity with n-grams: language-independent categorization of text. Science 267(5199):843–848. https://doi.org/10.1126/science.267.5199.843

    Article  CAS  PubMed  Google Scholar 

  42. Capecchi A, Probst D, Reymond J-L (2020) One molecular fingerprint to rule them all: drugs, biomolecules, and the metabolome. J Cheminf 12(1):43. https://doi.org/10.1186/s13321-020-00445-4

    Article  CAS  Google Scholar 

  43. Weininger D (1988) SMILES, a chemical language and information system. 1. introduction to methodology and encoding rules. J Chem Inf Comput Sci 28(1):31–36. https://doi.org/10.1021/ci00057a005

    Article  CAS  Google Scholar 

  44. Weininger D, Weininger A, Weininger JL (1989) SMILES. 2. algorithm for generation of unique smiles notation. J Chem Inf Comput Sci 29(2):97–101. https://doi.org/10.1021/ci00062a008

    Article  CAS  Google Scholar 

  45. Mendez D, Gaulton A, Bento AP, Chambers J, De Veij M, Félix E, Magariños MP, Mosquera JF, Mutowo P, Nowotka M, Gordillo-Marañón M, Hunter F, Junco L, Mugumbate G, Rodriguez-Lopez M, Atkinson F, Bosc N, Radoux CJ, Segura-Cabrera A, Hersey A, Leach AR (2019) ChEMBL: towards direct deposition of bioassay data. Nucleic Acids Res 47(D1):D930–D940. https://doi.org/10.1093/nar/gky1075

    Article  CAS  PubMed  Google Scholar 

  46. Sorokina M, Merseburger P, Rajan K, Yirik MA, Steinbeck C (2021) COCONUT Online: collection of open natural products database. J Cheminf 13(1):2. https://doi.org/10.1186/s13321-020-00478-9

    Article  Google Scholar 

  47. Irwin JJ, Tang KG, Young J, Dandarchuluun C, Wong BR, Khurelbaatar M, Moroz YS, Mayfield J, Sayle RA (2020) ZINC20—a free ultralarge-scale chemical database for ligand discovery. J Chem Inf Model. https://doi.org/10.1021/acs.jcim.0c00675

    Article  PubMed  PubMed Central  Google Scholar 

  48. Huang N, Shoichet BK, Irwin JJ (2006) Benchmarking sets for molecular docking. J Med Chem 49(23):6789–6801. https://doi.org/10.1021/jm0608356

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Rohrer SG, Baumann K (2009) Maximum unbiased validation (MUV) data sets for virtual screening based on pubchem bioactivity data. J Chem Inf Model 49(2):169–184. https://doi.org/10.1021/ci8002649

    Article  CAS  PubMed  Google Scholar 

  50. Blum LC, Reymond JL (2009) 970 million druglike small molecules for virtual screening in the chemical universe database GDB-13. J Am Chem Soc 131(25):8732–8733

    Article  CAS  PubMed  Google Scholar 

  51. Blum LC, van Deursen R, Reymond JL (2011) Visualisation and subsets of the chemical universe database GDB-13 for virtual screening. J Comput Aided Mol Des 25(7):637–647

    Article  CAS  PubMed  Google Scholar 

  52. McGinnis S, Madden TL (2004) BLAST: at the core of a powerful and diverse set of sequence analysis tools. Nucleic acids resea. https://doi.org/10.1093/nar/gkh435

    Article  Google Scholar 

  53. Dubikovskaya EA, Thorne SH, Pillow TH, Contag CH, Wender PA (2008) Overcoming multidrug resistance of small-molecule therapeutics through conjugation with releasable octaarginine transporters. Proc Natl Acad Sci USA 105(34):12128–12133. https://doi.org/10.1073/pnas.0805374105

    Article  PubMed  PubMed Central  Google Scholar 

  54. Stanzl EG, Trantow BM, Vargas JR, Wender PA (2013) Fifteen years of cell-penetrating, guanidinium-rich molecular transporters: basic science, research tools, and clinical applications. Acc Chem Res 46(12):2944–2954. https://doi.org/10.1021/ar4000554

    Article  CAS  PubMed  Google Scholar 

  55. Poirel L, Jayol A, Nordmann P (2017) Polymyxins: antibacterial activity, susceptibility testing, and resistance mechanisms encoded by plasmids or chromosomes. Clin Microbiol Rev 30(2):557–596. https://doi.org/10.1128/CMR.00064-16

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  56. Siriwardena TN, Gan B-H, Köhler T, van Delden C, Javor S, Reymond J-L (2021) Stereorandomization as a method to probe peptide bioactivity. ACS Cent Sci 7(1):126–134. https://doi.org/10.1021/acscentsci.0c01135

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  57. Buehler Y, Reymond J-L (2023) Molecular framework analysis of the generated database GDB-13s. J Chem Inf Model 63(2):484–492. https://doi.org/10.1021/acs.jcim.2c01107

    Article  CAS  PubMed  Google Scholar 

  58. Buehler Y, Reymond J-L (2023) Expanding bioactive fragment space with the generated database GDB-13s. J Chem Inf Model 63(20):6239–6248. https://doi.org/10.1021/acs.jcim.3c01096

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Funding

This work was supported by the Swiss National Science Foundation (200020_178998) and the European Research Council (885076).

Author information

Authors and Affiliations

  1. Department of Chemistry, Biochemistry and Pharmaceutical Sciences, University of Bern, Freiestrasse 3, 3012, Bern, Switzerland

    Markus Orsi & Jean-Louis Reymond

Authors
  1. Markus Orsi
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Jean-Louis Reymond
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

MO designed and realized the project and wrote the paper. JLR designed and supervised the project and wrote the paper. Both authors read and approved the final manuscript.

Corresponding author

Correspondence to Jean-Louis Reymond.

Ethics declarations

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1: Figure S1.

Mean and standard deviation of the pairwise ECFP4C similarities calculated for all 5 selected actives of each dataset contained in the benchmarking platform. Figure S2. Mean and standard deviation of the pairwise MAP4C similarities calculated for all 5 selected actives of each dataset contained in the benchmarking platform. Figure S3. Property distribution in the set uniformly sampled from the extended benchmark. Figure S4. Scatterplots of chiral shingle ratio vs. chiral atoms ratio. Figure S5. EF5 values across all small molecules and peptide targets. Figure S6. BEDROC20 values across all small molecules and peptide targets. Figure S7. BEDROC100 values across all small molecules and peptide targets. Figure S8. RIE20 valuesbvacross all small molecules and peptide targets. Figure S9. RIE100 values across all small molecules and peptide targets. Figure S10. Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked AUCs. Figure S11. Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked EF1s from benchmark datasets. Figure S12. Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked EF5s from benchmark datasets. Figure S13. a) Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked BEDROC20s from benchmark datasets. Figure S14. Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked BEDROC100s from benchmark datasets. Figure S15. Pairwise Pearson correlations andFriedman-Nemenyi test among tested fingerprints, based on the ranked RIE20s from benchmark datasets. Figure S16. Pairwise Pearson correlations and Friedman-Nemenyi test among tested fingerprints, based on the ranked RIE100s from benchmark datasets. Figure S17. Comparative analysis of MAP2C, MAP4C, MAP6C, APC, ECFP4C and ECFP6C Jaccard distance assignment on ln65 diastereomers and structural isomers. Figure S18. Comparative analysis of MAP2C, MAP4C, MAP6C, APC, ECFP4C and ECFP6C Jaccard distance assignment on polymyxin B2 diastereomers and structural isomers.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Orsi, M., Reymond, JL. One chiral fingerprint to find them all. J Cheminform 16, 53 (2024). https://doi.org/10.1186/s13321-024-00849-6

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s13321-024-00849-6

Keywords

Journal of Cheminformatics

ISSN: 1758-2946

Contact us