Data management in Medicinal Chemistry
OverviewQuestions:Objectives:
Why does medicinal chemistry research produce so much data?
How can big data be useful for medicinal chemists?
What are the current publicly available databases that can be used for drug discovery and development?
How to use Galaxy for data management and analysis of chemical data?
Requirements:
Learn some terminology from the field of medicinal chemistry
Understand the idea of data-driven medicinal chemistry
Explore the databases used for drug discovery and development
Use Galaxy tools for data management, format conversion and simple analyses
Time estimation: 1 hourLevel: Introductory IntroductorySupporting Materials:Published: Dec 23, 2024Last modification: Dec 23, 2024License: Tutorial Content is licensed under Creative Commons Attribution 4.0 International License. The GTN Framework is licensed under MITpurl PURL: https://gxy.io/GTN:T00429version Revision: 0
The development of medicinal chemistry is advancing very rapidly. Big pharmaceutical companies, research institutes and universities are working on ground-breaking solutions to help patients combat all kinds of diseases. During that development process, tons of data are generated – not only from the lab environment but also from clinical trials. Given that the discovery of more potent, safer and cheaper drugs is the ultimate goal of all research bodies, we should all focus on making the data we gather FAIR: Findable, Accessible, Interoperable, and Reusable to push the boundaries of drug development even further.
With the currently available methods such as artificial intelligence, machine learning, many toolkits, software and access to various databases, managing big data is now inherently linked to medicinal chemistry and helps to make this area as efficient as it can be.
In this tutorial, we will therefore explore some concepts related to medicinal chemistry, explore the available chemical and pharmacological databases, and perform some basic data analyses using the Galaxy interface.
AgendaIn this tutorial, we will cover:
Principles of medicinal chemistry research
Drug-likeness
There are many factors that medicinal chemists take into account while designing new drugs. It is essential to estimate the properties of the molecule before synthesising it in the lab, so the structures of drug candidates are usually compared to the existing drugs by so-called drug-likeness. This includes intrinsic properties of a compound that will lead to favourable ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) characteristics. Below are some properties usually assessed:
- Lipophilicity
- Size
- Polarity
- Insolubility
- Unsaturation
- Flexibility
Additionally, there have been some rules developed to help estimate drug-likeness, amongst which:
- Lipinski rule (Lipinski et al. 2001)
- Ghose rule (Ghose et al. 1999)
- Veber rule (Veber et al. 2002)
- Egan rule (Egan et al. 2000)
- Muegge rule (Muegge et al. 2001)
- Abbott bioavailability score (Martin 2005)
As you see, there are a number of characteristics to consider before even synthesising the molecule in a lab. One of the online tools that help to summarise all the most important information about intrinsic properties and drug-likeness of compounds is called SwissADME and is a free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules (Daina et al. 2017).
Below we briefly discuss two probably the most well-known rules: Lipinski’s and Veber’s rules.
The Lipinski rule of 5
This rule of thumb developed back in 1997 by Christopher Lipinski and colleagues tries to predict the likelihood that a given small molecule can be orally active. The Lipinski rule of 5 favours molecules as potential oral drug candidates if:
- the molecular mass is less than 500
- the calculated logarithm of the octanol−water partition coefficient (clogP) is less than 5
- they have up to 5 H-bond donors
- they have up to 10 H-bond acceptors
However, nowadays there are more and more drugs being developed which don’t comply with those rules and regardless are still effective. There are voices from the scientific community, pointing out that “We are in danger of repeating our past mistakes if we assume these new modalities are not ‘drug-like’ and cannot be oral drugs because they are not [rule of 5] compliant” (Michael Shultz from Novartis, cited in Mullard 2018 where the authors re-assess the rule of 5). Then, in O′Hagan et al. 2014 we read “This famous “rule of 5” has been highly influential in this regard, but only about 50 % of orally administered new chemical entities actually obey it.”
Veber’s rule
In (Veber et al. 2002) we read that the commonly applied molecular weight cutoff at 500 does not itself significantly separate compounds with poor oral bioavailability from those with acceptable values. The authors explain that on average both the number of rotatable bonds and polar surface area or hydrogen bond count tend to increase with molecular weight which may in part explain the success of the molecular weight parameter in predicting oral bioavailability. Their observations suggest that compounds will have a high probability of good oral bioavailability if they meet the two criteria:
- 10 or fewer rotatable bonds - increased rotatable bond count has a negative effect on the permeation rate
- a polar surface area no greater than 140 Å 2 (or 12 or fewer H-bond donors and acceptors) - reduced polar surface area correlates better with increased permeation rate than does lipophilicity
Big data for drug discovery
Have you ever wondered how new drugs are developed? It is actually quite a long and complex pathway and not only takes lots of time but also resources. The scheme below will help you understand the scale of that process.
As you can see, to get just one drug to the market, often thousands of structures are assessed to find the most effective and safe compound. That’s lots of information being generated in the whole process! The information can be stored in repositories and databases so that subsequent drug discovery process in a similar area is easier, faster and cheaper. Managing big data and using machine learning and computational chemistry methods is crucial in the lead optimisation step. By reviewing chemical databases and studying molecular docking simulations, we can save much time and resources to synthesise only those compounds that are the most promising based on in-silico methods. Even after synthesising the optimised ‘leads’, we still get lots of data from tests and analyses. By using appropriate cheminformatic tools and software, we can use the results to drive forward this iterative process of lead optimisation.
Ten Vs of big data for drug discovery
Hopefully, now you have an idea of what the process of drug discovery looks like. To understand why managing data in medicinal chemistry is so challenging, let’s have a closer look at the “Ten Vs” of big data Zhao et al. 2020:
- Volume: size of data.
- Velocity: speed of new data generation.
- Variety: various formats of data.
- Veracity: quality of data.
- Validity: authenticity of data.
- Vocabulary: terminology of data.
- Venue: platform of data generation.
- Visualization: view of data.
- Volatility: duration of data usefulness.
- Value: the potential of data usefulness to reduce the cost of drug discovery and development.
Medicinal chemists must keep in mind those features while both searching for data and publishing it. In this way we can focus on the main aspects of data management, try to improve the availability and normalisation of data, and be aware of limitations of repositories and inconsistencies in data quality.
FAIR MedChem
As you saw before, there are many properties that the compounds can be searched by, such as molecular weight, H-bond donors or acceptors, polarity, etc. Sometimes there is a need to assess other molecular properties but it might happen that not all the information is included in the database. This is one of the reasons why new repositories are being developed – they are more specific and gather particular properties of interest. How much easier the life of scientists could be if the relevant data was publicly available, well-ordered and contained the needed metadata? By submitting the data with the necessary information to the repository is a good way to make the data FAIR. This will make it:
- Findable, as the data will be given specific identifiers
- Accessible, as the data will be available online, open and free where possible
- Interoperable, as the repository will often enforce the use of formalised, consistent language
- Reusable, as the data will be released under a license with detailed provenance In the repositories
Chemical and pharmacological databases
Currently, there are lots of publicly available databases storing information about hit molecules, chemical structures and properties, drug targets, pharmaceuticals… In the recent paper Zhao et al. 2020 the authors collated relevant databases to advance computer-aided drug discovery and divided them into several main groups:
- Chemical collections
- Drug / drug-like compounds
- Drug targets, including genomics and proteomics data
- Biological data from assay screening, metabolism, and efficacy studies
- Drug liabilities and toxicities
- Clinical databases
Below you will find the databases listed under corresponding categories, all taken from Zhao et al. 2020 paper.
There are many more databases available, and many are still being developed. They are often quite specific and contain certain types of compounds (eg. PROTACs or are aimed at a particular disease. If the listed databases are not specific enough for your research, you can try luck by searching smaller, more specific databases.
Chemical collections
Database | Description | Size (as of 29 October 2019) |
---|---|---|
Enamine REAL Database | Tool used to find new hit molecules using large-scale virtual screening and for searching analogues to hit molecules | >700 million compounds that comply with ‘rule of 5’ and Veber criteria |
ZINC | Contains compound information including 2D/3D structure, purchasability, target, and biology-related information | >230 million compounds in 3D formats and >750 million compounds for analogue searching |
PubChem | Contains chemical molecule (mostly small molecule) information, including chemical structures, identifiers, chemical and physical properties, biological activities, safety and toxicity data | 97 million compounds, 236 million substances, 268 million bioactivities |
ChemSpider | Free chemical structure database providing fast access to >78 million structures, properties, and associated information | >78 million compound structures |
SCUBIDOO | Freely accessible database that currently holds 21 million virtual products originating from a small library of building blocks and collection of robust organic reactions | 21 million virtual products |
ChEMBL | Manually curated database of bioactive molecules with drug-like properties; brings together chemical, bioactivity, and genomic data to aid translation of genomic information into effective new drugs. | >1.9 million compounds, 1.1 million pieces of assay information |
TCM-Mesh | Integration of a database and a data-mining system for network pharmacology analysis of all respects of traditional Chinese medicine, including herbs, herbal ingredients, targets, related diseases, adverse effect, and toxicity | 383 840 compounds, 6235 herbs |
Super Natural II | Contains natural compounds, including information about corresponding 2D structures, physicochemical properties, predicted toxicity class and potential vendors | 325 508 natural compounds |
BIAdb | Comprehensive database of benzylisoquinoline alkaloids, containing information about 846 unique benzylisoquinoline alkaloids | ~846 unique benzylisoquinoline alkaloids |
Drug / drug-like compounds
Database | Description | Size (as of 29 October 2019) |
---|---|---|
AICD | Anti-Inflammatory Compounds Database (AICD) deposits compounds with potential antiinflammation activities | 79 781 small molecules |
DrugBank | Unique bioinformatics and cheminformatics resource that combines detailed drug data with comprehensive drug target information | 13 441 drug entries |
ReFRAME | Screening library of 12 000 molecules assembled by combining three databases (Clarivate Integrity, GVK Excelra GoStar, and Citeline Pharmaprojects) to facilitate drug repurposing | 12 000 molecules |
SuperDRUG2 | Contains approved/marketed drugs with regulatory details, chemical structures (2D and 3D), dosage, biological targets, physicochemical properties, external identifiers, adverse effects, and PK data | >4600 active pharmaceutical ingredients |
Drugs@FDA database | Information about drugs from FDA | ~23 391 drug application records |
e-Drug3D | Contains 1930 molecular structures approved by FDA between 1939 and 2019 with a molecular weight <2000 | 1930 drugs |
Drug targets, including genomics and proteomics data
Database | Description | Size (as of 29 October 2019) |
---|---|---|
BindingDB | Public, web-accessible database of measured binding affinities, focusing chiefly on interactions of proteins considered to be candidate drug targets with ligands that are small, drug-like molecules | 1 756 093 binding data, for 7371 protein targets and 780 240 small molecules |
Supertarget | An extensive web resource for analyzing drugtarget interactions. | 332 828 drug-target interactions |
Ligand Expo | Provides chemical and structural information about small molecules within structure entries of Protein Data Bank. | 30 440 entries of ligand |
PDBeChem | Consistent and enriched library of ligands, small molecules, and monomers referenced as residues and hetgroups in PDB entries | >29 922 ligands |
PDBbind-CN | Provides essential linkage between energetic and structural information of biomolecular complexes, which is helpful for various computational and statistical studies on molecular recognition in biological systems | 19 588 biomolecular complexes |
STITCH | Database integrating information about interactions from metabolic pathways, crystal structures, binding experiments, and drug–target relationships | Interactions between 300 000 small molecules and 2.6 million proteins from 1133 organisms |
BioGRID | The Biological General Repository for Interaction Datasets is an open-access database on protein, genetic, and chemical interactions for humans and all major model organisms | 1 753 686 protein and genetic interactions, 28 093 chemical associations and 874 796 posttranslational modifications from major model organisms |
Binding MOAD | Created from a subset of Protein Data Bank (PDB), containing every high-quality example of ligand-protein binding. | 36 047 protein-ligand structures, and 13 353 binding data |
GPCRdb | Contains data from GPCRs, including crystal structures, sequence alignments, and receptor mutations; can be visualized in interactive diagrams; provides online analysis tools | 15 149 proteins, and 144 917 ligands |
Guide to Pharmacology | IUPHAR/BPS guide to pharmacology is an open-access, expert-curated database of molecular interactions between ligands and their targets. | 2937 targets, and 9859 ligands |
GLASS | GPCR-Ligand Association (GLASS) database is a manually curated repository for experimentally validated GPCR–ligand interactions; along with relevant GPCR and chemical information, GPCRligand association data are extracted and integrated into GLASS from literature and public databases | ~ 277 651 unique ligands and 3048 GPCRs |
Biological data from assay screening, metabolism, and efficacy studies
Database | Description | Size (as of 29 October 2019) |
---|---|---|
HMDB | Freely available electronic database containing detailed information about small-molecule metabolites found in the human body | 114 162 metabolite entries |
SMPDB | Small Molecule Pathway Database (SMPDB) is an interactive, visual database containing >30 000 small-molecule pathways found in humans only | >30 000 small-molecule pathways |
TTD | Therapeutic Target Database (TTD) is a database providing information about known and explored therapeutic protein and nucleic acid targets, targeted disease, pathway information and corresponding drugs directed at each of these targets | 2589 targets, and 31 614 drugs |
BioCyc | Collection of 7615 pathway/genome databases; each database in BioCyc collection describes genome and metabolic pathways of a single organism | 7615 pathway/genome databases |
BiGG | Metabolic reconstruction of human metabolism designed for systems biology simulation and metabolic flux balance modelling | 2004 proteins, 2766 metabolites, and 3311 metabolic and transport reactions |
BRENDA | Main collection of enzyme functional data available to scientific community | At least 40 000 different enzymes from >6900 different organisms |
Reactome | Curated, peer-reviewed knowledgebase of biological pathways, including metabolic pathways, and protein trafficking and signalling pathways | >9600 proteins, 9800 reactions, and 2000 pathways for humans |
BioModels Database | Repository of computational models of biological processes; models described from literature are manually curated and enriched with cross-references | 6753 patient-derived genome-scale metabolic models, 112 898 metabolic models etc. |
KEGG | Database resource that integrates genomic, chemical, and systemic functional information | 18 652 metabolites |
CARLSBAD | Database and knowledge inference system that integrates multiple bioactivity data sets to provide researchers with novel capabilities for mining and exploration of available structure-activity relationships (SAR) throughout chemical biology space. | 932 852 CARLSBAD activities, 890 323 unique structure–target pairs, 3542 targets, 435 343 unique structures |
WOMBAT | Contains 331 872 entries, representing 1966 unique targets, with bioactivity annotations | 268 246 unique structures |
Open NCI Database | Maintained by the National Cancer Institute; contains small-molecule information such as names, biological activities, and structures; a useful resource for researchers working in cancer/AIDS fields | >250 000 compounds |
NPACT | Provides information on the plant-derived natural compounds, including structure, properties (physical, elemental, and topological), cancer type, cell lines, inhibitory values (IC50, ED50, EC50, GI50), molecular targets, commercial suppliers, and drug-likeness of compounds | 1574 entries |
PKPB_DB | Contains physiological parameter values for humans from early childhood through senescence; intended to be used in physiologically based (PB) PK modelling; also contains similar data for animals (primarily rodents) | N/A |
Drug liabilities and toxicities
Database | Description | Size (as of 29 October 2019) |
---|---|---|
T3DB | Unique bioinformatics resource that combines detailed toxin data with comprehensive toxin target information | 3678 toxins |
DrugMatrix | One of world’s largest toxicogenomic reference resources | ~ 600 drug molecules and 10 000 genes |
ACToR | Includes computational toxicology information about compounds, including HTS, chemical exposure, sustainable chemistry (chemical structures and physicochemical properties) and virtual tissue data | >500 000 chemicals |
SkinSensDB | Contains curated data from published AOP-related skin sensitization assays | 710 unique chemicals |
SIDER | Contains information on marketed medicines and their recorded adverse drug reactions, including frequency, drug and adverse effect classifications | 1430 drugs with 5868 side effect information |
LTKB Benchmark Dataset | Contains drugs with potential to cause drug-induced liver injury in humans; established using FDA-approved prescription drug labels; liver-toxicity-knowledge-base-ltkb/ltkb-benchmark-dataset | 287 prescription drugs |
CTD | Comparative Toxicogenomics Database (CTD) is a premier public resource for literature-based, manually curated associations between chemicals, gene products, phenotypes, diseases, and environmental exposures | 13 378 unique chemicals and related information |
Clinical databases
Database | Description | Size (as of 29 October 2019) |
---|---|---|
ClinicalTrials.gov | Database of privately and publicly funded clinical studies conducted around the world | ~ 324 429 research studies in all 50 US states and 209 countries |
AACT database | Publicly available relational database that contains all information (protocol and result data elements) about every study registered in ClinicalTrials.gov. Content is downloaded from ClinicalTrials.gov daily and loaded into AACT | ~ 324 429 research studies in all 50 US states and 209 countries |
EORTC Clinical Trials Database | Contains information about EORTC clinical trials and clinical trials from other organizations with EORTC participation | N/A |
Exposome-Explorer | Contains detailed information on the nature of biomarkers, populations and subjects where measured, samples analyzed, methods used for biomarker analyses, concentrations in biospecimens, correlations with external exposure measurements, and biological reproducibility over time | 908 biomarkers |
PharmaGKB | A pharmacogenomics knowledge resource that encompasses clinical information about drug molecules | 733 drugs with their clinical information |
Data management in chemistry using Galaxy
Chemical file formats
Depending on the type of analysis you are going to perform, you will work with different file formats. Below are the most important data types, commonly used in cheminformatics, that you will likely see in Galaxy:
- SDF (.sd, .sdf) - stands for Structure-Data Format, and SDF files wrap the molfile (MDL Molfile) format. Stores information about the chemical structure and associated data of compounds in plain text. Files in SDF format can encode single or multiple molecules that are then delimited by lines consisting of four dollar signs. SDF files are formatted ASCII files that store information about the positions of the individual atoms (either in 2D or 3D space) that make up the molecule. The data on connectivity and hybridization state are also encoded, although their use is less frequent and often inconsistent.
- MOL (.mol) - an MDL Molfile for holding information about the atoms, bonds, connectivity and coordinates of a molecule.
- MOL2 (.mol2) - a Tripos Mol2 file is a complete, portable representation of a SYBYL molecule. It is an ASCII file which contains all the information needed to reconstruct a SYBYL molecule
- CML (.cml) - Chemical Markup Language (ChemML or CML) is an approach to managing molecular information using tools such as XML and Java. It supports a wide range of chemical concepts including molecules, reactions, spectra and analytical data, computational chemistry, chemical crystallography and materials
- InChI (IUPAC International Chemical Identifier) - a textual identifier for chemical substances, designed to provide a standard way to encode molecular information. The identifiers describe chemical substances in terms of layers of information — the atoms and their bond connectivity, tautomeric information, isotope information, stereochemistry, and electronic charge information (Heller et al. 2015)
- SMILES (.smi) - the simplified molecular-input line-entry system (SMILES) is a specification in the form of a line notation for describing the structure of chemical species using short ASCII strings. A linear text format which can describe the connectivity and chirality of a molecule (Weininger 1988)
- PDB - the Protein Data Bank (PDB) file format is a textual file format describing the three-dimensional structures of molecules held in the Protein Data Bank, now succeeded by the mmCIF format. It contains a description and annotation of protein and nucleic acid structures including atomic coordinates, secondary structure assignments, as well as atomic connectivity. In addition, experimental metadata is stored. (Berman 2007)
- GRO (.gro) - a plain text file storing spatial coordinates and velocities (if available) of atoms during a molecular dynamics simulation, utilised by GROMACS
To visualise the structures held by the files with the positions of the individual atoms, you can use (NGL Viewer)[https://nglviewer.org/ngl/].
If you plan to work with molecular dynamics simulations, there are also some MD trajectory file formats that you might want to get familiar with. In Galaxy, you can convert between xtc, trr, dcd and netcdf files using a tool called MDTraj file converter ( Galaxy version 1.9.6+galaxy0).
Galaxy tools
In Galaxy Chemical Toolbox there are dozens of tools that can be used for various analyses. Below we will only show a few, mostly related to data import, format conversion and some functions linked to what was discussed previously.
Let’s start with importing publicly available data. Protein Data Bank stores thousands of three-dimensional structural data of proteins nucleic acids and other biological molecules.
Hands-on: Import PDB file from Protein Data Bank
- Open Protein Data Bank
- Type your search term in the box “Enter search term(s), Entry ID(s), or sequence”:
MAO B
- Choose the structure of interest and copy its accession code: 2BK3
- Switch to Galaxy
- Get PDB file ( Galaxy version 0.1.0) with the following parameters:
- “PDB accession code”:
2BK3
This is often the first step of docking studies that you can learn about in other tutorials, for example in Protein-ligand docking tutorial.
You can import any molecule to Galaxy using SMILES notation. Below we show an example with benzenesulfonyl chloride and ethylamine.
Hands-on: Import SMILES
- Copy SMILES of your molecule(s) of interest. In this example, we use benzenesulfonyl chloride (SMILES: C1=CC=C(C=C1)S(=O)(=O)Cl) and ethylamine (SMILES: CCN).
- Click on the galaxy-upload Upload data button in the tools panel on the left-hand side.
- Click on the galaxy-wf-edit Paste/Fetch data button twice. Two boxes will appear.
- In the first box, under “Name” section, enter
Benzenesulfonyl chloride
and under “Type”:smi
- Below paste the SMILES of benzenesulfonyl chloride:
C1=CC=C(C=C1)S(=O)(=O)Cl
- In the second box, under “Name” section, enter
Ethylamine
and under “Type”:smi
- Below paste the SMILES of ethylamine:
CCN
- Click the “Start” button, then close the dialogue box. Your files are being added to your history!
As you learned from the previous section, there are many different formats used in computational chemistry. Of course, Galaxy allows you to interconvert between them - just have a look below!
Hands-on: Convert the file format
- Compound conversion ( Galaxy version 3.1.1+galaxy0) with the following parameters:
- param-files “Molecular input file”: click on param-files Multiple datasets icon and choose both
Benzenesulfonyl chloride
andEthylamine
- “Output format” - you have many options to choose from! Not only the formats but also the associated parameters. Just scroll the list and pick the one that is relevant to your downstream analysis. Here we use
MDL MOL format (sdf, mol)
- Rename galaxy-pencil the corresponding files
Benzenesulfonyl chloride SDF
andEthylamine SDF
. Before renaming the files, make sure to check either the starting dataset numbers or the atoms in the file in order not to confuse the two files!
We intentionally chose to work on benzenesulfonyl chloride and ethylamine. Do you know why? Well, if you think about the synthesis of sulfonamides, that’s exactly what you need - a sulfonyl chloride and an amine. So… let’s do some computational synthesis!
Hands-on: Run the reaction
- Reaction maker ( Galaxy version 1.1.4+galaxy0) with the following parameters:
- param-file “Input file”:
Benzenesulfonyl chloride SDF
- “Reagent options”: Sulfonamide
- param-file “Reagent file”:
Ethylamine SDF
Above we converted the .smi file into .sdf to show how tool Compound conversion tool works. However, it is worth pointing out that tool Reaction maker works also with SMILES files and the conversion happens automatically so that your .smi dataset is imported as .sdf input.
tool Reaction maker works with more reaction types, such as:
- Amides synthesis
- Nucleophilic aromatic substitution (SNAr mechanism)
- Urea
- Suzuki coupling
- Sulfonamide
- Reductive amination
- N-alkylation
- Ether coupling
- Ester coupling
- Benzimidazole
- Triazole
- Benzoxazole
- Sarah Cu
- Sarah Quat Am
The output is the .txt log file and the SDF file with the product molecule. If you have a look at that file, you’ll see that there is just a list of atoms with their corresponding coordinates. To draw the resulting structure, we can use another tool.
Hands-on: Visualisation of compounds
- Visualisation ( Galaxy version 3.1.1+galaxy0) with the following parameters:
- param-file “Molecular input file”:
SDF output for Reaction maker
- “Property to display under the molecule” - there are many properties that you can choose from! Please note that sometimes if not enough information is provided, then the property might not be displayed. Let’s choose a basic but useful property:
Molecular weight
- “Format of the resulting picture”:
SVG
Since we spoke about the drug-likeness in the previous section, let’s see how it works in practice, on the example of our “synthesised” molecule.
Hands-on: Estimate the drug-likeness
- Drug-likeness ( Galaxy version 2021.03.4+galaxy0) with the following parameters:
- param-file “Molecule data in SDF or SMILES format”:
SDF output for Reaction maker
- “Method” - two possible methods to weight the features are available - one based on max. weight (QEDw,max), and the other on mean weight (QEDw,mo). There is also an option to leave features unweighted (QEDw,u) and we’ll choose this one:
unweighted (QEDw,u)
- “Include the descriptor names as header”: param-toggle
Yes
The eight properties used are molecular weight (MW), octanol–water partition coefficient (ALOGP), number of hydrogen bond donors (HBDs), number of hydrogen bond acceptors (HBAs), molecular polar surface area (PSA), number of rotatable bonds (ROTBs), number of aromatic rings (AROMs) and number of structural alerts (ALERTS).
In medicinal chemistry, we often base the new structures on scaffolds of existing drugs. Therefore, it is quite useful to be able to find similar structures - below is an example of how to do it in Galaxy.
Hands-on: Search ChEMBL database for similar compounds
- Search ChEMBL database ( Galaxy version 0.10.1+galaxy4) with the following parameters:
- “SMILES input type”:
File
- param-file “Input file”:
Benzenesulfonyl chloride
- “Search type”:
Substructure
- Rename galaxy-pencil the output file
Benzenesulfonyl chloride substructures
Even though the input format of the above tool is SMILES, it can automatically interconvert between .sdf and .smi formats, so you don’t have to use tool Compound conversion tool beforehand.
tool Search ChEMBL database tool allows for searching the ChEMBL database for compounds which resemble a SMILES string. Two search options are possible:
- similarity (searches for compounds which are similar to the input within a specified Tanimoto cutoff)
- substructure (searches for compounds which contain the input substructure)
Results can be filtered for compounds which are
- approved drugs
- biotherapeutic
- natural products
- fulfil all of the Lipinski rule of five criteria
Let’s repeat this step, but with Lipinski’s Rule of Five which you learned about in the previous section of this tutorial!
Hands-on: Search ChEMBL database for similar compounds using the Lipinski rule of five
- Click the galaxy-refresh Repeat button on the previous dataset.
- You will be redirected to Search ChEMBL database ( Galaxy version 0.10.1+galaxy4). Leave all the parameters as before, except for one:
- “Filter for Lipinski’s Rule of Five”: param-toggle
Yes
- Rename galaxy-pencil the output file
Benzenesulfonyl chloride Lipinski substructures
QuestionHow many molecules have been filtered out?
If you click on the dataset, you will see a short summary of how many molecules have been found. When we didn’t apply any filters, the tool found 45 molecules. After applying Lipinski’s Rule of Five filter, 36 molecules satisfied the rules.
Galaxy workflow summary
Those are just a few tools, but they offer many parameters that you can tune depending on your analysis. Additionally, we provide you with an example history and the dedicated workflow of this small tutorial. You are more than welcome to explore other tools in the Galaxy Chemical Toolbox though!
Data-driven medicinal chemistry - yesterday, today and tomorrow
Why do we need big data in med-chem?
In the article by Lusher et al. 2014 the role of the medicinal chemist is to make decisions about which of the infinite possibilities of new compounds should be made next. As the amount and variety of data on which to base these decisions grows, so too must the data analysis skills of the medicinal chemist. The authors claim that modern medicinal chemists should be able to recognise sources of relevant information, prepare raw data, use statistical tools, extract meaningful information, interpret results, recognise potential problems and make visualisations to communicate their findings to improve the quality of compounds being produced in research. According to the authors, this situation will require improved education and increased access to data and information management tools. This is where Galaxy comes in, offering not only a platform for data analysis with a plethora of the most commonly used tools in the field but also educational materials allowing anyone to learn and excel in analysis.
The privilege of applying computational methods in medicinal chemistry pipelines allows to focus on the most promising compounds and to remove unsuitable ones before the stage of chemical synthesis. Not only is this approach more efficient but also more sustainable and can reduce the costs of the synthetic stage of drug discovery (Brown et al. 2018).
Another important role that big data plays in drug discovery is target identification (and validation) which might be currently one the biggest challenges in medicinal chemistry. By introducing genomic data in drug discovery, we can produce more specific and effective medicines. Understanding the underlying causes of the disease and the biological targets are crucial in designing new drugs and hence should be inherently linked with this process, where possible.
Data challenges
It is important to understand the limitations of the databases we use since if data is not well understood, the processing and analysis drawn may ultimately be flawed, following the ‘garbage in, garbage out’ principle. Here are some challenges mentioned in Brown et al. 2018 that you can explore more in-depth by referring to that article.
- Errors
- Reproducibility
- Standardisation
- Formatting of data
- Information held in silos
- Loss of contextual information
- Limited and biased data
Machine learning in med-chem
Given the huge number of molecules in the various databases, applying machine learning in cheminformatics and building predictive models of the physiochemical properties of molecules is more and more popular. It helps to predict absorption, distribution, metabolism, excretion, toxicity (ADMET), likelihood of interaction with the drug targets as well as off-target effects. It can be useful in elucidating complex protein-protein or drug-drug interaction networks, in understanding structure-activity relationships (SARs) and when combined with data from the ‘omics’ revolution. This approach also helps to identify bioisosteres and introduce the idea of scaffold hopping in molecular design (Brown et al. 2018).
Knowledge sharing
It is not only about gathering and managing the data but also building a culture of knowledge sharing. We work in big, international teams with well-established methods and well-equipped labs. There are so many opportunities for knowledge capture and exchange! The challenge emerges here though – how to efficiently connect and help each other? This is the question that many pharmaceutical companies try to answer, such as Merck & Co. who describe their approach to knowledge management and report on the multiple enduring and complementary teams and initiatives to capture and share knowledge (Beshore et al. 2022).