Chapter 2: Ligand-based Screening: Filtering & Similarity Searching
Introduces virtual screening as a computational way to prioritize promising compounds for focused experimental testing.
After this chapter you will be able to
- Virtual and experimental screening. Distinguish virtual screening from experimental high-throughput screening, including how experimental data can support computational screening.
- Enriched compound prioritization. Explain what virtual screening selects from a chemical library and why enrichment matters for subsequent testing.
- Virtual Screening’s Initial Aim. Explain that the immediate aim is to screen a compound library computationally for hits with promising antimalarial activity, not to produce preclinical-ready drugs in one step.
- Hierarchical screening pipeline. Explain why virtual screening commonly starts with cheap, broad steps and reserves more intensive methods for the compounds that remain.
- Target structure method choice. Distinguish structure-based from ligand-based screening by the information available about the target.
- Known ligand evidence levels. Explain which ligand-based approaches become possible with one active compound, several actives, or many labeled active and inactive compounds.
- Screening library source trade-off. Distinguish using an existing purchasable catalog from building a custom or enumerated library, and explain the trade-off between chemical-space coverage and synthetic feasibility.
- Known actives as screening references. Explain how this ligand-based screening scenario uses known antimalarial compounds to assess a diverse candidate library.
- Reference Based Similarity Search. Trace the basic similarity-searching strategy: use a compound with a known activity or property as a reference, rank library compounds by similarity to it, and retain promising matches for experimental validation.
- Interdependent Search Design Choices. Describe the three choices that complicate a similarity search—molecular representation, similarity metric, and search strategy—and explain that the chosen features affect comparisons and search results.
- Limits of Structural Similarity. Explain why a structurally similar compound is not guaranteed to have the desired function: different parts of a molecule may matter, be irrelevant, or work against that function.
- Dataset molecules as RDKit objects. Describe how the SDF dataset is loaded into a Pandas DataFrame and how its SMILES can be converted to RDKit Mol objects. Distinguish a Mol object from a picture of a molecule.
- Connection tables and molecular graphs. Explain how an atom table and a bond table specify a molecule as a graph, and why different connection tables can represent the same molecule.
- SDF information and entry blocks. Describe what SDFs add to basic connection tables and identify the structural information recorded in an example SDF entry’s blocks.
- Chemical validity and sanitization. Explain that sanitization checks whether a SMILES is chemically reasonable before conversion to a Mol, and why this matters for reliable screening data.
- Canonical and isomeric SMILES. Explain why SMILES are canonicalized to reduce duplicate structures, how retaining chiral and isotopic information affects what counts as redundant, and what Chem.MolToSmiles does by default.
- Purpose of molecular descriptors. Explain why SMILES strings are processed into molecular descriptors and how those descriptors support filtering and property prediction.
- Purpose and types of filters. Explain why compound filters are used in a large screening library and distinguish property-based filters from substructure-based filters by what each considers.
- Oral Absorption and Polarity Balance. Explain why an orally administered drug needs a balance of water solubility and membrane permeability, unlike an IV drug that enters the circulation directly.
- Rule of Five Criteria. State Lipinski’s four Rule of Five criteria and explain how lipophilicity and hydrogen-bonding limits relate to oral absorption risk.
- PAINS and BRENK Filters. Explain why PAINS and BRENK substructure filters are used when screening a compound library, and distinguish the problems each flags.
- Structural Alert Limitations. Explain how toxicity-related structural alerts are used and why an alert is not a definitive toxicity prediction.
- Custom SMARTS Substructure Filtering. Describe how the Glaxo Wellcome hard filters illustrate a custom substructure filter, including what SMARTS patterns and RDKit’s HasSubstructMatch() contribute.
- Binary structural fingerprints. Explain how a molecular fingerprint represents a molecule’s two-dimensional structural features as a fixed-length vector of 1s and 0s, rather than as properties such as molecular weight.
- Fingerprint-based compound screening. Describe how fingerprints can be used to find library molecules structurally similar to known desirable compounds, and why the fingerprint type depends on the structural information needed.
- Structural Key Bit Encoding. Explain how a structural-key fingerprint uses a predefined fragment dictionary to assign features to bit positions and sets each bit according to whether a molecule contains its corresponding feature.
- Fixed Dictionary Limitations. Explain why structural keys cannot represent features missing from their fragment dictionary and why the dictionary should suit the intended molecules and use case.
- Hashed Fingerprint Construction. Explain why hashed fingerprints are an alternative to structural keys and describe how a path-based fingerprint is built.
- Hash Collisions in Screening. Explain why hash collisions occur, how assigning multiple bits per fragment limits their effect, and what kind of screening error collisions can cause.
- Circular Fingerprint Neighborhoods. Distinguish circular hashed fingerprints from path-based fingerprints and relate Morgan radius 2 to ECFP4.
- Binary Versus Count Fingerprints. Distinguish binary from count-based Morgan fingerprints and explain when counts may be useful.
- Morgan fingerprint settings and tradeoffs. Explain why the library uses radius-2 Morgan fingerprints of length 2048, how radius 2 differs from radius 4, and why fingerprint settings should be reported.
- Feature-based similarity searching. Explain how similarity searching uses the features of an item of interest to find similar items in a collection, even when they are not identical.
- Tanimoto and Dice calculations. Describe how Tanimoto and Dice similarity quantify overlap between sets or binary fingerprints, including each coefficient’s denominator and the meaning of scores from 0 to 1.
- Conservative Tanimoto scoring. Explain why Dice tends to produce a higher score than Tanimoto for the same molecular pair and why the section favors Tanimoto when searching for close analogs of active compounds.
- Tanimoto and Dice Scores. Explain how Tanimoto and Dice behaved when the library was compared with one Malaria Box query, and why the same numerical cutoff would yield different hit counts.
- Similarity Threshold Limitations. Explain why a similarity threshold cannot be treated as a universal guarantee of biological activity, including the effects of molecular features and activity cliffs.
- Early Screening Recall Choice. Justify the section’s choice of Dice similarity with a 0.65 cutoff for early screening, and contrast it with a situation favoring Tanimoto.
What it builds on
- Target Identification & Hit Discovery (chapter 1) · The role of finding hits in drug discovery.
- What is a Molecule? The Joy of SMILES (chapter 1) · The reader has encountered SMILES as a way to represent molecules.
- The Challenge of Drug Discovery (chapter 1) · the drug-discovery context for pharmacokinetic properties
Section by section
Introduction
Introduces virtual screening as a computational way to prioritize promising compounds for focused experimental testing.
- Virtual and experimental screening: Distinguish virtual screening from experimental high-throughput screening, including how experimental data can support computational screening.
Why it matters: This explains why computational screening can extend the reach of laboratory screening. - Enriched compound prioritization: Explain what virtual screening selects from a chemical library and why enrichment matters for subsequent testing.
Why it matters: The aim is not just to rank compounds, but to make follow-up experiments more productive and less costly.
Key terms: virtual screening, high-throughput screening, machine learning models, chemical library, target, enriched subset, active compounds, random selection
2.1 What is Virtual Screening?
The section frames virtual screening a compound library for promising antimalarial hits as an initial step toward a broader drug-discovery pipeline.
- Virtual Screening’s Initial Aim: Explain that the immediate aim is to screen a compound library computationally for hits with promising antimalarial activity, not to produce preclinical-ready drugs in one step.
Why it matters: This distinction keeps the first screening task separate from the longer-term goal of producing optimized candidates.
Key terms: virtual screening, compound library, hits, optimized drug candidates
2.1.1 Virtual Screening Taxonomy
This section explains how virtual screening narrows a compound library and how the available target structure and known compounds guide the choice of screening method.
- Hierarchical screening pipeline: Explain why virtual screening commonly starts with cheap, broad steps and reserves more intensive methods for the compounds that remain.
Why it matters: The order of a pipeline determines where its computational resources are spent. - Rule of Five filter: Describe what Lipinski’s Rule of Five checks and the limited prediction it is used to make.
Why it matters: This gives a concrete example of a fast compound filter and what its result means. - Contextual filter decisions: Explain why drug-likeness filters should be treated as guidelines and how flagged compounds might be handled.
Why it matters: Treating a filter result as a verdict could discard promising candidates. - Target structure method choice: Distinguish structure-based from ligand-based screening by the information available about the target.
Why it matters: The available evidence determines which family of screening methods is feasible. - Known ligand evidence levels: Explain which ligand-based approaches become possible with one active compound, several actives, or many labeled active and inactive compounds.
Why it matters: Machine learning is one option among ligand-based methods, not the default for every dataset.
Key terms: hierarchical virtual screening, compound filters, ligand-based virtual screening, molecular docking, molecular dynamics simulations, Lipinski’s Rule of Five, drug-likeness, molecular weight, calculated logP, hydrogen bond acceptors
2.1.2 Scenario: Hit Identification of Antimalarial Compounds
This section explains why antimalarial hit discovery is needed, how screening libraries are chosen, and why known active compounds provide a reference for ligand-based screening.
- Antimalarial screening rationale: Explain why the search for new antimalarial candidates remains important, including the threat of drug resistance in P. falciparum.
Why it matters: Drug resistance explains why finding new antimalarial candidates is an ongoing need. - Screening library source trade-off: Distinguish using an existing purchasable catalog from building a custom or enumerated library, and explain the trade-off between chemical-space coverage and synthetic feasibility.
Why it matters: A useful screening library must offer promising chemical diversity without losing a practical path to testing its compounds. - Known actives as screening references: Explain how this ligand-based screening scenario uses known antimalarial compounds to assess a diverse candidate library.
Why it matters: This identifies what the candidates are being compared against and why structural resemblance is relevant to the screen.
Key terms: malaria, Plasmodium falciparum, resistance, screening libraries, commercial catalog, enumerated library, synthetic feasibility, chemical-space coverage, ligand-based virtual screening, SPECS
2.1.3 Strategy: Similarity Searching
This section introduces similarity searching from a known compound, identifies the choices that shape a search, and explains why structural similarity does not guarantee similar activity.
- Reference Based Similarity Search: Trace the basic similarity-searching strategy: use a compound with a known activity or property as a reference, rank library compounds by similarity to it, and retain promising matches for experimental validation.
Why it matters: This is the starting strategy that the chapter’s later screening steps refine. - Interdependent Search Design Choices: Describe the three choices that complicate a similarity search—molecular representation, similarity metric, and search strategy—and explain that the chosen features affect comparisons and search results.
Why it matters: A similarity ranking only makes sense in light of the features, comparison method, and search strategy used to produce it. - Limits of Structural Similarity: Explain why a structurally similar compound is not guaranteed to have the desired function: different parts of a molecule may matter, be irrelevant, or work against that function.
Why it matters: This limitation explains why a close structural match is a candidate for investigation, not proof of the desired effect.
Key terms: reference compound, similarity searching, library, experimental validation, representation, features, similarity metric, search strategy, structural similarity, functionality
2.2 Loading a Virtual Screening Library
The section previews loading a molecular dataset, checking molecule quality, and calculating descriptors.
2.2.1 Understanding the Dataset as a Structure Data File
Explains how to load the screening dataset as RDKit molecules and how an SDF represents molecular structures and associated information.
- Dataset molecules as RDKit objects: Describe how the SDF dataset is loaded into a Pandas DataFrame and how its SMILES can be converted to RDKit Mol objects. Distinguish a Mol object from a picture of a molecule.
Why it matters: The DataFrame and Mol objects provide the molecular data used for subsequent processing. - Connection tables and molecular graphs: Explain how an atom table and a bond table specify a molecule as a graph, and why different connection tables can represent the same molecule.
Why it matters: Connection tables explain the structural foundation on which SDF entries build. - SDF information and entry blocks: Describe what SDFs add to basic connection tables and identify the structural information recorded in an example SDF entry’s blocks.
Why it matters: Recognizing the blocks makes it easier to interpret what the loaded file contains beyond a list of molecules.
Key terms: Structure Data File (SDF), Pandas DataFrame, PandasTools, SMILES, RDKit Mol object, connection table, atom table, bond table, molecule as a graph, bond order
2.2.2 Molecule Sanitization
This section explains how to validate molecular inputs and standardize SMILES so a screening dataset contains chemically sensible, nonredundant structures.
- Chemical validity and sanitization: Explain that sanitization checks whether a SMILES is chemically reasonable before conversion to a Mol, and why this matters for reliable screening data.
Why it matters: Invalid molecular inputs can undermine the reliability of screening results. - Use-dependent validity decisions: Distinguish the section’s usual reliance on RDKit’s default validation from additional use-dependent decisions about which compounds to retain, and describe when sanitization might be disabled.
Why it matters: A chemically valid input is not necessarily suitable for every screening purpose. - Canonical and isomeric SMILES: Explain why SMILES are canonicalized to reduce duplicate structures, how retaining chiral and isotopic information affects what counts as redundant, and what Chem.MolToSmiles does by default.
Why it matters: Consistent SMILES representations help prevent repeated structures from adding noise to the dataset.
Key terms: sanitization, SMILES, Mol, valence rules, RDKit, default validation and sanitization, sanitize=False, non-standard chemistry, canonicalize, unique SMILES
2.2.3 Molecular Descriptors
The section explains why molecules are represented with numerical descriptors and identifies four RDKit descriptors computed for the screening library.
- Purpose of molecular descriptors: Explain why SMILES strings are processed into molecular descriptors and how those descriptors support filtering and property prediction.
Why it matters: Descriptors make molecular information usable for the compound filters that follow. - Computed screening descriptors: Describe the four descriptors computed with RDKit for each molecule and their immediate role in the screening dataset.
Why it matters: These are the specific properties available for the next stage of compound filtering.
Key terms: SMILES, molecular descriptors, featurize, RDKit, molecular weight, hydrogen bond acceptors, hydrogen bond donors, logP
2.3 Compound Filters
Introduces compound filtering as a low-cost way to reduce a screening library and distinguishes property-based from substructure-based filters.
- Purpose and types of filters: Explain why compound filters are used in a large screening library and distinguish property-based filters from substructure-based filters by what each considers.
Why it matters: This distinction clarifies what information each kind of filter uses before applying it to a screening library.
Key terms: compound filters, screening library, property-based filters, descriptors, substructure-based filters, substructures
2.3.1 Property-based Filters
Explains how Lipinski’s Rule of Five filters a compound library for properties associated with oral absorption, and shows the filter’s effect on the library.
- Oral Absorption and Polarity Balance: Explain why an orally administered drug needs a balance of water solubility and membrane permeability, unlike an IV drug that enters the circulation directly.
Why it matters: This trade-off explains why a property filter considers both solubility and membrane crossing. - Rule of Five Criteria: State Lipinski’s four Rule of Five criteria and explain how lipophilicity and hydrogen-bonding limits relate to oral absorption risk.
Why it matters: The criteria give a practical way to flag compounds whose properties may make oral absorption harder. - Filtered Library Changes: Describe how applying the Ro5 property filter changed the SPECS compound library and its reported property distributions.
Why it matters: Filtering changes the makeup of the library used for subsequent screening.
Key terms: absorption, bioavailability, polarity, Lipinski’s Rule of Five, Ro5, molecular weight, logP, hydrogen bond acceptors, hydrogen bond donors, property filter
2.3.2 Structure-based Filters
This section explains how substructure filters remove likely assay-interfering or otherwise undesirable compounds, how to apply a custom SMARTS filter, and why filter results need careful interpretation.
- PAINS and BRENK Filters: Explain why PAINS and BRENK substructure filters are used when screening a compound library, and distinguish the problems each flags.
Why it matters: These filters help remove compounds likely to waste screening effort for different reasons. - Structural Alert Limitations: Explain how toxicity-related structural alerts are used and why an alert is not a definitive toxicity prediction.
Why it matters: An alert is useful for triage, but treating it as a verdict can discard useful molecules or overlook risks. - Custom SMARTS Substructure Filtering: Describe how the Glaxo Wellcome hard filters illustrate a custom substructure filter, including what SMARTS patterns and RDKit’s HasSubstructMatch() contribute.
Why it matters: Custom patterns let you screen for liabilities beyond RDKit’s predefined filters. - Filter Counts and Tradeoffs: Explain why the number of compounds removed by a filter depends on earlier filters, and why removal counts should be checked against screening efficiency and potential losses.
Why it matters: Removal counts are meaningful only in the context of the library and the filters already applied.
Key terms: PAINS, BRENK filters, substructure-based filters, false positive hits, structural alerts, toxic endpoints, user-defined threshold, REOS, SMARTS, HasSubstructMatch()
2.4 Fingerprints: Representing Molecules as Numbers
Introduces molecular fingerprints as fixed-length binary representations of structure and explains how they support virtual screening by identifying compounds similar to a query.
- Binary structural fingerprints: Explain how a molecular fingerprint represents a molecule’s two-dimensional structural features as a fixed-length vector of 1s and 0s, rather than as properties such as molecular weight.
Why it matters: This representation lets you describe what the later similarity search compares. - Fingerprint-based compound screening: Describe how fingerprints can be used to find library molecules structurally similar to known desirable compounds, and why the fingerprint type depends on the structural information needed.
Why it matters: It connects the binary representation to the chapter’s search for compounds worth investigating.
Key terms: molecular fingerprints, fragmental descriptors, bit vector, structural features, virtual screening, similarity searching, query, structural keys, hashed fingerprints
2.4.1 Structural Keys
Explains how structural-key fingerprints encode predefined molecular features as bits and why their fixed feature dictionary limits what they can represent.
- Structural Key Bit Encoding: Explain how a structural-key fingerprint uses a predefined fragment dictionary to assign features to bit positions and sets each bit according to whether a molecule contains its corresponding feature.
Why it matters: This is how a molecule's structure becomes a bit vector that can be compared with others. - Fixed Dictionary Limitations: Explain why structural keys cannot represent features missing from their fragment dictionary and why the dictionary should suit the intended molecules and use case.
Why it matters: A fingerprint cannot supply a model with information about features it never encodes.
Key terms: fingerprint bit vectors, structural keys, fragment dictionary, structural feature, structural features, MACCS keys
2.4.2 Hashed Fingerprints
This section explains how path-based and circular hashed fingerprints represent molecules, what information they can lose, and when count-based fingerprints may help.
- Hashed Fingerprint Construction: Explain why hashed fingerprints are an alternative to structural keys and describe how a path-based fingerprint is built.
Why it matters: This explains how varied molecular fragments become a representation that can be compared without maintaining a fixed fragment dictionary. - Hash Collisions in Screening: Explain why hash collisions occur, how assigning multiple bits per fragment limits their effect, and what kind of screening error collisions can cause.
Why it matters: It clarifies why a fingerprint match may need further screening without implying that collisions hide genuine matches. - Circular Fingerprint Neighborhoods: Distinguish circular hashed fingerprints from path-based fingerprints and relate Morgan radius 2 to ECFP4.
Why it matters: The distinction makes clear which molecular patterns each fingerprint type captures. - Fingerprint Representation Limits: Explain how fingerprint parameters can make different molecules indistinguishable and why that matters for modeling.
Why it matters: Identical representations can conceal differences a property model needs to distinguish. - Binary Versus Count Fingerprints: Distinguish binary from count-based Morgan fingerprints and explain when counts may be useful.
Why it matters: The choice determines whether repeated molecular features influence a similarity comparison.
Key terms: structural keys, path-based hashed fingerprints, fragment dictionary, hash function, fixed-length fingerprint, collision, substructure, false positive, false negative, circular hashed fingerprints
2.4.3 Fingerprinting our Library
The section explains the Morgan fingerprint settings used for the library, why radius 2 was chosen, and how to inspect the molecular fragments represented by fingerprint bits.
- Morgan fingerprint settings and tradeoffs: Explain why the library uses radius-2 Morgan fingerprints of length 2048, how radius 2 differs from radius 4, and why fingerprint settings should be reported.
Why it matters: The chosen settings determine which molecular details the library fingerprints emphasize and let others reproduce the results. - Fingerprint bit fragment visualization: Describe how a molecule’s set fingerprint bits can be visualized as fragments and what the radius-2 limit means in the example.
Why it matters: Viewing the fragments makes the local structural information encoded by a fingerprint easier to interpret.
Key terms: Morgan (circular) fingerprints, radius, fingerprint length, fingerprint bits, fragments, central atom
2.5 Similarity Searching
Introduces similarity searching as comparing an item’s features with those of a collection to find items that resemble it without being identical.
- Feature-based similarity searching: Explain how similarity searching uses the features of an item of interest to find similar items in a collection, even when they are not identical.
Why it matters: This gives you the basic idea behind searching a collection for promising items that resemble one you already care about.
Key terms: similarity searching, features
2.5.1 Defining “Similarity”
Defines molecular similarity using fingerprint features and explains how Tanimoto and Dice coefficients measure and compare fingerprint overlap.
- Similarity in fingerprint space: Explain how molecular fingerprints let us compare molecules as points in a feature space, where closer representations indicate greater similarity in the represented features.
Why it matters: This gives you an intuitive way to think about what a molecular similarity score compares. - Tanimoto and Dice calculations: Describe how Tanimoto and Dice similarity quantify overlap between sets or binary fingerprints, including each coefficient’s denominator and the meaning of scores from 0 to 1.
Why it matters: The formulas tell you what a fingerprint similarity score counts—and what it cannot prove. - Conservative Tanimoto scoring: Explain why Dice tends to produce a higher score than Tanimoto for the same molecular pair and why the section favors Tanimoto when searching for close analogs of active compounds.
Why it matters: Choosing a coefficient affects which molecules appear sufficiently similar to an active compound.
Key terms: fingerprints, fingerprint feature space, similarity, Tanimoto similarity, Jaccard similarity, Dice similarity, intersection, union, binary vectors, close analogs
2.5.2 Searching against a Query
This section applies a fingerprint-based similarity search to the Malaria Box queries, explains how metric and threshold choices affect hits, and narrows the results for downstream screening.
- Tanimoto and Dice Scores: Explain how Tanimoto and Dice behaved when the library was compared with one Malaria Box query, and why the same numerical cutoff would yield different hit counts.
Why it matters: A cutoff only makes sense in relation to the metric that produces the scores. - Similarity Threshold Limitations: Explain why a similarity threshold cannot be treated as a universal guarantee of biological activity, including the effects of molecular features and activity cliffs.
Why it matters: A similarity hit is a candidate for further assessment, not evidence that it will have the query’s biological effect. - Early Screening Recall Choice: Justify the section’s choice of Dice similarity with a 0.65 cutoff for early screening, and contrast it with a situation favoring Tanimoto.
Why it matters: The search settings reflect whether missing a promising compound or retaining extra candidates is the bigger concern. - Query Hits and Budget: Describe how the search defines a hit across Malaria Box queries and how it prioritizes matches when downstream screening has a fixed budget.
Why it matters: The hit rule and budget determine which candidates advance from a large library to further screening.
Key terms: Tanimoto similarity, Dice similarity, similarity threshold, descriptors, features, activity cliffs, Dice coefficient, Tanimoto, recall, 0.65 cutoff
2.6 Summary
Where this is used later
- Enrichment Factors in Virtual Screening (chapter 8) · Provides the introductory meaning of enrichment in virtual screening.
- (chapter intro) (chapter 9) · Introduces virtual screening of compound libraries.
- Docking: A Core SBDD Technique (chapter 9) · Assumes the earlier introduction to virtual screening.
- Protein-Ligand Docking (chapter 9) · Provides the target-structure context for protein–ligand docking.
- Your Machine Learning Task (chapter 3) · Frames a machine-learning task using compound examples.
- Learning from Data (chapter 3) · Learning from labeled data offers a way to identify which structural features matter across molecules with known activity.
- Loading and Exploring the hERG Blockers Dataset (chapter 3) · Working with a dataset of molecular structures.
- Prepare the Protein & Ligand Structures (chapter 9) · Prior work with molecular structure files provides context for extracting structures from a PDB file.
Concepts explained in this chapter
- similarity searchSimilarity search retrieves or ranks molecules according to their structural similarity to one or more query molecules.
- compound filterCompound filtering is the computational removal of molecules with undesirable properties or substructures from a screening collection.
- Lipinski's rule of fiveThe Lipinski rule of five is a heuristic set of molecular-property guidelines for estimating a compound's oral drug-likeness.
- drug-likenessDrug-likeness is the degree to which a compound has physicochemical and structural characteristics associated with suitability as a drug.
- ligand-based virtual screeningLigand-based virtual screening prioritizes compounds by using information from known ligands rather than an explicit target structure.
- virtual screeningVirtual screening is computational prioritization of chemical compounds as drug candidates by predicting target interactions or drug-relevant properties.
- structure-property relationshipA structure-property relationship describes how molecular structure corresponds to a chemical, biological, or functional property.
- pipelineA sequential pipeline is an architecture in which distinct processing stages execute in order and pass each stage's outputs to the next stage.
- Dice similarity coefficientDice similarity is a metric that compares two molecular fingerprints using twice the number of shared features divided by the sum of their feature counts.
- Jaccard similarityTanimoto similarity is a metric that compares two sets or binary molecular fingerprints as the ratio of shared features to their union.
- ADMETAn ADMET property describes a drug compound's absorption, distribution, metabolism, excretion, or toxicity behavior in a biological system.
- QSAR modelQSAR modeling predicts molecular properties or biological activities from quantitative relationships between chemical structure and measured outcomes.
- hierarchical virtual screeningHierarchical virtual screening is a staged screening strategy that progressively applies more complex and computationally expensive methods to a shrinking compound set.
- ligand-based machine learningLigand-based machine learning predicts compound activity from relationships between molecular structural features and observed activity data.
- machine learningMachine learning is a computational approach that learns patterns or predictive functions from data rather than relying solely on explicitly programmed rules.
- molecular dockingMolecular docking is computational simulation of how a molecule fits and orients within a target binding site to estimate its binding behavior.
- pharmacophore-based screeningPharmacophore-based sampling selects compounds that match hypothesized spatial and chemical features required for target binding.
- structure-based drug designStructure-based drug design is a drug-design approach that uses three-dimensional molecular structures and interactions to guide compound discovery and optimization.
- molecular fingerprintA molecular fingerprint is a fixed-length, often binary, representation encoding structural features or substructures present in a molecule.
- Morgan fingerprintA Morgan fingerprint is a circular molecular fingerprint that encodes atom-centered structural neighborhoods within a specified radius.
- hashed fingerprintA path-based hashed fingerprint encodes linear molecular paths by hashing them into a fixed-length bit representation.
- similarity metricA similarity coefficient is a numerical measure that quantifies the resemblance between two sets or binary representations.
- BRENK filterA BRENK filter is a compound filter that identifies unwanted structural groups associated with reactivity, metabolic instability, toxicity, or other drug-development liabilities.
- Glaxo Wellcome hard filterThe Glaxo Wellcome hard filter is a rule-based substructure filter for removing compounds with undesirable structural motifs from chemical libraries.
- PAINSA PAINS compound is a pan-assay interference compound whose structure is associated with nonspecific or misleading activity in biological assays.
- PAINS filterA PAINS filter is a substructure-based filter that flags chemical motifs associated with pan-assay interference and unreliable activity measurements.
- REOSREOS is a compound-filtering concept for rapidly removing low-quality molecules likely to be unproductive or produce assay artifacts.
- SMARTSSMARTS is a chemical pattern language extending SMILES to represent molecular substructural queries.
- structural alertA structural alert is a chemical substructure associated with undesirable reactivity, toxicity, instability, or other drug-development risk.
- activity cliffAn activity cliff is a pair or series of structurally similar molecules that exhibit a large difference in biological activity.
- molecular descriptorA molecular descriptor is a numerical or symbolic representation of a molecule's physicochemical, structural, topological, or quantum-mechanical properties.
- compound libraryA compound library is a curated collection of chemical compounds available for screening or analysis.
- molecule objectAn RDKit Mol is an in-memory molecular graph object representing atoms, bonds, and associated molecular information.
- hydrogen bond acceptorA hydrogen-bond acceptor is an electronegative atom or functional group that can accept a hydrogen bond from a donor.
- hydrogen bond donorA hydrogen-bond donor is an electronegative atom or functional group bearing a hydrogen that can donate a hydrogen bond.
- lipophilicityLogP is the logarithm of a compound's octanol-water partition coefficient and is a measure of lipophilicity.
- drug candidateA clinical candidate is a drug-development compound selected for advancement into clinical testing based on its efficacy, safety, and development properties.
- feature spaceA feature space is a mathematical space in which objects are represented as points defined by measurable features.
- embeddingAn embedding is a learned numerical vector representation of an object that captures properties relevant to a task.
- canonical SMILESA unique SMILES is a canonicalized SMILES representation selected so that equivalent molecular structures receive the same textual representation.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
