Drug Discovery

Drug discovery is the multidisciplinary scientific pipeline encompassing target identification, hit generation, hit-to-lead expansion, multiparameter lead optimization, and preclinical safety qualification through which novel therapeutic candidates are created and prepared for human clinical investigation. Situated at the intersection of medicinal chemistry, structural biology, pharmacology, chemoinformatics, and translational medicine, modern drug discovery transforms a causal hypothesis about a human disease mechanism into a clinical candidate possessing balanced pharmacodynamics (target affinity and selectivity) and pharmacokinetics (ADMET and formulation stability).

Developing a single approved new molecular entity (NME) typically requires 10 to 15 years of research and development, screening from thousands to billions of initial chemical structures, and navigating an overall clinical-stage attrition rate of approximately 88%–90% between Phase I first-in-human studies and regulatory approval. To reduce late-stage clinical failures driven by lack of efficacy (~50% of Phase II/III failures) and safety liabilities (~25%), contemporary discovery integrates human genomic validation, structural biology (X-ray crystallography, cryo-electron microscopy, and AlphaFold), and predictive computational design.

Target-Based versus Phenotypic Drug Discovery

Modern pharmaceutical discovery operates through two complementary strategic frameworks analyzed extensively by David Swinney and Jason Anthony (2011):

• Target-Based Drug Discovery (TDD): Begins with a defined molecular hypothesis linking a single purified macromolecular target—such as a mutated oncogenic kinase (tyrosine kinase inhibitors), a G-protein-coupled receptor, or a viral protease—to disease pathophysiology. High-throughput biochemical or biophysical screens then identify ligands that modulate that specific target.
• Phenotypic Drug Discovery (PDD): Screens chemical libraries directly in disease-relevant cellular, organoid, or whole-organism models (such as patient-derived induced pluripotent stem cells or bacterial cultures) to identify compounds that reverse a pathological phenotype without prior assumptions regarding the molecular target. Target deconvolution is performed subsequently via phenotypic screening and chemoproteomics, cellular thermal shift assays (CETSA), or resistance mutation sequencing.

Genomic Target Validation and Human Genetic Evidence

Because half of all clinical Phase II trials fail due to insufficient clinical efficacy—indicating that modulating the chosen molecular target did not alter the human disease course—rigorous target identification and validation is the earliest gate in the discovery pipeline. Experimental validation combines loss-of-function and gain-of-function perturbations using genome-wide CRISPR-Cas9 knockout, CRISPRi transcriptional repression, or RNA interference screens paired with single-cell RNA-Seq transcriptomics.

Human population genetics provides the strongest causal validation in drug discovery. Retrospective analyses by Matthew Nelson and colleagues (2015) and Emily King and colleagues (2019) demonstrated that drug targets supported by direct human genetic associations—from Genome-Wide Association Studies (GWAS) or rare Mendelian loss-/gain-of-function variants in biobanks—are more than twice as likely to progress from Phase I clinical trials to regulatory approval. A landmark example is PCSK9 (proprotein convertase subtilisin/kexin type 9): the discovery that rare gain-of-function PCSK9 mutations cause autosomal dominant hypercholesterolemia while loss-of-function nonsense variants confer lifelong 30%–40% reductions in LDL cholesterol and an 88% reduction in coronary heart disease directly inspired the monoclonal antibody therapy agents evolocumab and alirocumab and the GalNAc-conjugated siRNA inclisiran (antisense oligonucleotide and siRNA therapeutics).

High-Throughput Screening (HTS) and Assay Quality (Z'-Factor)

Once a target is validated, hit identification searches chemical space (including synthetic small molecules and natural product drug discovery) for initial active molecules ("hits") with reproducible micromolar to sub-micromolar activity (IC₅₀ or K_d typically 0.1–10 μM). In automated High-Throughput Screening (HTS), robotic liquid handlers test corporate compound decks of 10⁵ to 10⁶ diverse small molecules in 384- or 1536-well plates using fluorescence polarization, Förster resonance energy transfer (TR-FRET), luminescence, or RapidFire mass spectrometry readouts.

Assay robustness is quantified statistically prior to screening using the Z'-factor formulated by Ji-Hu Zhang, Thomas Chung, and Kevin Oldenburg (1999):

*Z*' = 1 − [3 · (σ_p + σ_n)] / |μ_p − μ_n|

where μ_p and μ_n are the mean signals and σ_p and σ_n are the standard deviations of the positive (100% inhibition) and negative (DMSO vehicle) controls. A *Z*' ≥ 0.5 indicates an excellent screening assay with a wide separation band between signal and noise. Primary HTS hits are subsequently triaged through orthogonal counter-screens to eliminate Pan-Assay Interference Compounds (PAINS) (such as redox-cycling quinones or rhodanines) and colloidal aggregators (detected by adding non-ionic detergent Triton X-100).

Fragment-Based Drug Discovery (FBDD), DELs, and Virtual Screening

Three major technological platforms complement conventional HTS:

1. Fragment-Based Drug Discovery (FBDD): Screens small libraries (1,000–3,000 compounds) of low-molecular-weight polar "fragments" obeying the Rule of Three (MW ≤ 300 Da, Clog*P* ≤ 3, H-bond donors ≤ 3, H-bond acceptors ≤ 3). Because small fragments form fewer contacts, their initial affinities are weak (K_d ≈ 100 μM to 10 mM) and require biophysical detection via Surface Plasmon Resonance (SPR) and Isothermal Titration Calorimetry (ITC), 2D protein-observed NMR, or high-throughput X-ray crystallography soaking. Co-crystal structures then guide fragment growing, merging, or linking into nanomolar leads—a strategy that yielded the approved oncology drugs vemurafenib (BRAF^V600E inhibitor), venetoclax (BCL-2 protein-protein interaction inhibitor), and sotorasib (KRAS^G12C inhibitor via covalent drug design).
2. DNA-Encoded Chemical Libraries (DELs): Developed from concepts by Sydney Brenner and Richard Lerner (1992), DELs use split-and-pool combinatorial synthesis to attach a unique double-stranded DNA barcode to every building-block step, creating libraries of 10⁹ to 10¹² distinct compounds in a single microcentrifuge tube. After affinity capture against an immobilized protein target, bound ligands are identified via PCR amplification and massive parallel sequencing.
3. Structure-Based Drug Design (SBDD) and Virtual Screening: Uses molecular docking, pharmacophore modeling, generative AI in drug discovery, and alchemical free energy perturbation (FEP+) across ultra-large make-on-demand virtual libraries (such as Enamine REAL, exceeding 10¹⁰ compounds) to prioritize top-scoring molecules for synthesis.

Ligand Efficiency (LE) and Lipophilic Ligand Efficiency (LLE)

During hit-to-lead and lead optimization, medicinal chemists iteratively execute the Design–Make–Test–Analyze (DMTA) cycle—employing bioisosterism and scaffold hopping as well as peptide and macrocycle therapeutics—to explore Structure–Activity Relationships (SAR) while conforming to physicochemical guidelines such as Lipinski's rule of five. Because adding hydrophobic atoms to a scaffold frequently inflates molecular weight and lipophilicity—creating "molecular obesity" that degrades solubility and increases CYP3A4 clearance and hERG cardiotoxicity—Andrew Hopkins, Colin Groom, and Alexander Alex (2004) introduced Ligand Efficiency (LE), which normalizes standard binding free energy (Δ*G*°) by the number of non-hydrogen heavy atoms (N_heavy):

LE = −Δ*G*° / N_heavy = (−*R* · T · ln K_d) / N_heavy ≈ (1.37 · p*IC*₅₀) / N_heavy (in kcal·mol⁻¹·atom⁻¹)

To reach a 1 nM oral drug (pIC₅₀ = 9, Δ*G*° ≈ −12.3 kcal/mol) with a molecular weight below 500 Da (N_heavy ≤ 36–38), a chemical series must maintain LE ≥ 0.30 kcal·mol⁻¹·heavy atom⁻¹ from fragment hit through final candidate. Equally important is Paul Leeson and Benito Springthorpe's (2007) Lipophilic Ligand Efficiency (LLE, or LipE):

LLE = p*IC*₅₀ (or p*K*_d) − cLog*P*

Optimizing leads toward LLE ≥ 5–7 (for example, p*IC*₅₀ = 8.5 with cLog*P* = 2.5) ensures that nanomolar potency is driven by directional polar interactions (hydrogen bonds and salt bridges) rather than non-specific hydrophobic partitioning.

Preclinical GLP Toxicology, IND Filing, and Clinical Trial Phases

Once a single optimized clinical candidate demonstrates robust in vivo pharmacokinetics–pharmacodynamics (PK/PD) target engagement in disease models, it enters Investigational New Drug and CMC development. Solid-state pharmaceutics and Biopharmaceutics Classification System (BCS) formulation or targeted drug delivery systems and lipid nanoparticles optimize bioavailability. Under International Council for Harmonisation (ICH M3/S7A/S7B) guidelines, the candidate undergoes Chemistry, Manufacturing, and Controls (CMC) scale-up, genotoxicity testing (Ames and micronucleus assays), core safety pharmacology, and preclinical toxicology and safety pharmacology in two mammalian species—one rodent (rat or mouse) and one non-rodent (beagle dog or non-human primate). The No-Observed-Adverse-Effect Level (NOAEL) in the most sensitive species is combined with physiologically based pharmacokinetic (PBPK) modeling and allometric scaling to establish the Maximum Recommended Starting Dose (MRSD) for human clinical trial phases: Phase I (safety, tolerability, and human PK in 20–100 subjects), Phase II (proof-of-concept efficacy and dose-ranging in 100–300 patients), Phase III (confirmatory randomized controlled trials in 1,000–5,000 patients for New Drug Application and FDA regulatory approval), and Phase IV post-marketing pharmacovigilance, followed post-exclusivity by bioequivalence, generics, and biosimilars and systematic drug repurposing and network pharmacology.

Why do fragment-based drug discovery (FBDD) libraries require only ~2,000 compounds compared to >500,000 compounds in high-throughput screening (HTS)?

According to chemical space combinatorics (estimated by Jean-Louis Reymond in the GDB databases), the number of possible drug-like molecules grow by roughly 8-fold for every heavy atom added: there are ~10⁶⁰ theoretical molecules at 30 heavy atoms (MW ~450 Da), but only ~1.4 × 10⁷ theoretical molecules at ≤11 heavy atoms (MW ~160 Da). Furthermore, as shown by Mike Hann's molecular complexity model (2001), small polar fragments (MW ≤ 300 Da) have a much lower probability of encountering steric clashes inside a binding pocket than larger HTS molecules. Consequently, 1,500–2,000 diverse fragments sample binding sub-pockets far more efficiently than a million complex HTS compounds, provided sensitive biophysical methods (X-ray, NMR, SPR) are used to detect their weak initial K_d values (0.1–5 mM).

How do Ligand Efficiency (LE) and Lipophilic Ligand Efficiency (LLE) prevent medicinal chemists from falling into "molecular obesity"?

During lead optimization, adding lipophilic aromatic rings or alkyl chains easily improves in vitro potency (IC₅₀) via the hydrophobic effect, but simultaneously inflates molecular weight and Log*P*, precipitating poor aqueous solubility, rapid CYP3A4 metabolism, and off-target hERG blockade (ADMET). Tracking Ligand Efficiency (LE = 1.37 · p*IC*₅₀ / N_heavy ≥ 0.30) ensures every added heavy atom contributes at least 0.30 kcal/mol of binding free energy, while tracking Lipophilic Ligand Efficiency (LLE = p*IC*₅₀ − cLog*P* ≥ 5.0) forces chemists to gain potency without increasing lipophilicity—confirming that added functional groups make specific, geometrically matched hydrogen-bonding or electrostatic contacts with the protein target.

Why are two animal species (one rodent and one non-rodent) required for GLP preclinical toxicology before Phase I human trials?

Regulatory agencies (FDA, EMA, PMDA under ICH M3(R2)) mandate repeat-dose GLP toxicology in both a rodent species (rat or mouse) and a non-rodent species (dog, minipig, or non-human primate) because species-specific differences in Phase I and Phase II drug metabolism, plasma protein binding, biliary excretion, and target receptor homology can either mask or exaggerate organ toxicities in a single species. Evaluating concordance across two evolutionarily divergent mammals identifies target organs of toxicity (liver, kidney, bone marrow, heart) and establishes a conservative No-Observed-Adverse-Effect Level (NOAEL) to protect Phase I volunteers.

1. Swinney, D. C., & Anthony, J. (2011). "How Were New Medicines Discovered?" Nature Reviews Drug Discovery, 10(7), 507–519. https://doi.org/10.1038/nrd3480
2. Nelson, M. R., Tipney, H., Painter, J. L., et al. (2015). "The Support of Human Genetic Evidence for Approved Drug Indications." Nature Genetics, 47(8), 856–860. https://doi.org/10.1038/ng.3314
3. Hopkins, A. L., Groom, C. R., & Alex, A. (2004). "Ligand Efficiency: A Useful Metric for Lead Selection." Drug Discovery Today, 9(10), 430–431. https://doi.org/10.1016/S1359-6446(04)03069-703069-7)
4. Erlanson, D. A., Fesik, S. W., Hubbard, R. E., Jahnke, W., & Jhoti, H. (2016). "Twenty Years On: The Impact of Fragments on Drug Discovery." Nature Reviews Drug Discovery, 15(9), 605–619. https://doi.org/10.1038/nrd.2016.109