How Are Medications Discovered?

Every medication in the pharmacopoeia has a story. Some were born from laboratory accidents that changed history. Others emerged from systematic searches through millions of compounds. Today, artificial intelligence algorithms predict drug targets and design novel molecules with unprecedented speed. Yet despite technological revolution, the fundamental challenge remains unchanged: finding a safe, effective treatment for human disease.

For medical students and practicing clinicians, understanding how medications are discovered provides crucial context. It explains why certain drugs cost what they do, why the development timeline spans a decade or more, and why breakthrough innovations matter so profoundly. This exploration traces the journey from Fleming's moldy petri dish to cutting-edge AI platforms that may fundamentally reshape pharmaceutical innovation.

The Stories Behind the Breakthroughs: Five Discoveries That Changed Medicine

The most transformative medications often emerged not from careful planning, but from keen observation, scientific curiosity, and the willingness to follow unexpected leads. These five discoveries illustrate how different pathways can converge on life-saving treatments.

Modern Drug Discovery: Methods, Technologies, and Real-World Examples

Contemporary medication discovery employs an increasingly sophisticated arsenal of methodologies. While serendipity still plays a role, modern drug developers use systematic approaches to identify and optimize new compounds. Understanding these methods contextualizes current pharmaceutical innovation and helps explain why certain drugs reach market while others never progress beyond the laboratory.

1. Classical Pharmacology (Forward Pharmacology)

Principle: Researchers identify a biological target (receptor, enzyme, protein), then systematically screen compounds to find those that bind to and modulate the target.

Process: Classical approaches begin with defining a disease mechanism. For example, researchers might hypothesize that a particular receptor is dysregulated in a given condition. They then synthesize or source compounds designed to affect that receptor. These are screened in cell-based or tissue-based assays to identify "hits"—compounds showing the desired activity. Hits are then refined through chemistry and structure-activity relationship (SAR) analysis to improve potency, selectivity, and drug-like properties.

Historical Example: The discovery of SSRIs in the 1970s exemplified classical pharmacology. Based on the monoamine hypothesis of depression (that insufficient serotonin caused depression), researchers at multiple companies synthesized compounds designed to inhibit serotonin reuptake. Eli Lilly's fluoxetine (Prozac) emerged from this rational screening. The compound was selective for serotonin transporters, had favorable pharmacokinetics, and proved effective in clinical trials. Its 1987 FDA approval revolutionized psychiatric practice.

Current Use: Classical pharmacology remains a cornerstone of drug discovery. Many modern medications result from this systematic approach. However, it depends heavily on correctly identifying the disease-causing target—a challenge when disease biology is incompletely understood.

2. Serendipitous Discovery and Clinical Observation

Principle: Unexpected findings during clinical trials, basic research, or routine clinical practice lead to discovery of novel therapeutic uses.

Process: Clinicians and researchers maintain careful vigilance for unexpected effects. When observations diverge from predictions, rigorous investigation may reveal new therapeutic mechanisms or applications. This requires organizational cultures that reward unexpected findings and scientific teams prepared to pivot toward promising leads.

Modern Examples: Minoxidil was developed as an oral antihypertensive but was abandoned when blood pressure control proved inadequate. However, clinical observers noted that patients developed hirsutism (excess hair growth). Upjohn pursued this side effect, eventually reformulating minoxidil as a topical solution. It became the first FDA-approved medication for male pattern baldness (Rogaine). Similarly, botulinum toxin was developed as a treatment for strabismus (eye misalignment), but ophthalmologists noticed patients' wrinkles improved. This observation led to cosmetic and therapeutic applications that now drive a multibillion-dollar industry.

Limitations: Serendipitous discovery becomes rarer as pharmaceutical companies face pressure for predictable, high-return research programs. Yet fostering cultures that value unexpected observations remains important for breakthrough innovations.

3. High-Throughput Screening (HTS)

Principle: Rapidly test thousands to millions of compounds against a biological target to identify "hits" with desired activity.

Process: Researchers prepare a purified target (often a recombinant protein or cell line expressing the target). Robotic systems dispense compounds and measure binding or functional responses using automated assays. Modern HTS can evaluate 100,000+ compounds daily. Hits are then validated and optimized.

Advantages: HTS dramatically accelerates early-stage discovery by rapidly narrowing down massive libraries. It reduced the timeline from target identification to lead compounds from years to months. It enables discovery of compounds that would be unlikely through rational design alone.

Real-World Example: Many of GSK's (GlaxoSmithKline) marketed drugs, including asthma medications like zafirlukast (Accolate), emerged from HTS campaigns against leukotriene receptors. HTS identified the initial hits; subsequent chemistry refined these into drugs with superior properties.

Limitations: HTS identifies compounds that bind targets, not necessarily compounds that will be efficacious or safe in humans. Many hits fail to translate into clinical candidates because they lack drug-like properties (poor absorption, rapid metabolism, toxicity). The field evolved to address this through focus on "drug-likeness" criteria (Lipinski's Rule of Five).

4. Reverse Pharmacology (Target-Based Drug Design)

Principle: Start with a disease-associated target (identified through genetics, cell biology, or disease pathology), design compounds to modulate it, then test whether modulation provides therapeutic benefit.

Process: This approach relies on understanding disease biology sufficiently to identify a tractable target. Researchers then use structure-based design, computational modeling, or screening to identify compounds affecting the target. Only after identifying potent, selective modulators do researchers test efficacy in disease models and ultimately clinical trials.

Advantages: Reverse pharmacology is rational and hypothesis-driven. It can efficiently explore targets that might not be discovered through serendipity. It's particularly powerful when disease genetics clearly point to a target.

Example: The discovery of PCSK9 inhibitors (evolocumab, alirocumab) exemplified reverse pharmacology. Genetic studies identified PCSK9 loss-of-function variants associated with dramatically lower LDL cholesterol and reduced cardiovascular disease. This identified PCSK9 inhibition as a therapeutic target. Companies then designed monoclonal antibodies and small molecules to inhibit PCSK9, resulting in a new class of cholesterol-lowering drugs.

Limitations: Reverse pharmacology assumes the identified target is disease-causing. When disease mechanisms are misunderstood or multifactorial, this approach may lead to dead ends despite targeting "correct" biology.

5. Medicinal Chemistry and Structure-Activity Relationships (SAR)

Principle: Systematically modify compound structures to understand which chemical features drive desired activity, toxicity, and pharmacokinetic properties.

Process: Starting with an initial "hit" or "lead" compound, chemists synthesize structural analogs—variations with single or multiple chemical modifications. Each analog is tested for activity, selectivity, potency, and drug-like properties. Patterns emerge: this functional group improves potency; that substitution reduces toxicity; another modification improves oral bioavailability. Through iterative synthesis and testing, chemists optimize compounds toward clinical candidates.

Example: The evolution of antipsychotics demonstrates SAR. First-generation drugs like chlorpromazine were potent dopamine antagonists but caused severe extrapyramidal side effects. SAR investigations of related compounds led to second-generation antipsychotics (risperidone, olanzapine) with improved pharmacological profiles—rapid dissociation from dopamine receptors and additional serotonin antagonism reduced extrapyramidal effects while maintaining efficacy.

Timeline Impact: SAR optimization can consume 3-5 years and synthesize hundreds to thousands of compounds. Modern chemistry has accelerated this through parallel synthesis, where multiple analogs are prepared simultaneously rather than sequentially.

6. Computer-Aided Drug Design (CADD) and Molecular Docking

Principle: Use computational models of drug targets and molecular modeling software to predict which compounds will bind effectively and possess favorable properties.

Process: If the three-dimensional crystal structure of a target protein is known (often determined through X-ray crystallography or cryo-EM), computational chemists can model how small molecules fit into the target's binding pocket. Software predicts binding affinity, identifies probable binding poses, and highlights chemical features likely to enhance or diminish binding. This guides synthetic chemists in designing the next generation of compounds more efficiently.

Advantages: CADD accelerates lead optimization by prioritizing synthesis efforts toward the most promising compounds. It reduces the number of compounds that must be synthesized and tested empirically. Modern structure-based design has cut years off traditional drug discovery timelines.

Industrial Example: Many large pharmaceutical companies maintain dedicated computational chemistry divisions. GSK, Merck, and others use CADD extensively in optimization phases to predict ADME properties (absorption, distribution, metabolism, excretion) and off-target toxicities.

Limitations: CADD predictions are most accurate for enzymatic targets and well-characterized receptor binding. For complex cellular processes or poorly characterized targets, computational models remain less predictive. The "black box" nature of some computational methods can make predictions difficult to interpret.

7. Artificial Intelligence and Machine Learning in Drug Discovery

Principle: Train machine learning algorithms on vast datasets of molecular properties, target interactions, and bioactivity to predict novel compounds with desired characteristics.

Process: AI systems learn patterns from historical data (chemical structures, binding affinity, toxicity, pharmacokinetic properties, clinical outcomes). These models can then predict properties of novel compounds without synthesizing or testing them. Deep learning architectures, including convolutional neural networks and graph neural networks, excel at molecular property prediction. Some systems can generate novel molecular structures predicted to have desired properties.

Key Companies:

• Recursion Pharmaceuticals: Founded in 2014, Recursion merged with Exscientia in 2024 to combine phenotypic screening with AI-driven drug design. The company focuses on AI-predicted compounds tested in automated microscopy-based assays. Their approach aims to accelerate early discovery and reduce failure rates in later stages.
• Insilico Medicine: Founded in 2014, Insilico specializes in AI for target identification, lead generation, and biomarker discovery. Their generative models can design novel molecules predicted to bind targets or avoid off-target effects. In 2021, they reported designing a novel DDR1 kinase inhibitor in approximately 40 days—a process historically requiring years.
• Atomwise: Developed AI platforms for virtual screening against protein structures. Their AtomNet technology predicts binding interactions using deep learning trained on millions of historical docking results.

Successes and Promise: Several AI-designed compounds have entered clinical trials. Insilico's INS018_055, an AI-designed DDR1 kinase inhibitor, entered Phase 2 trials for fibrosis. Exscientia's EXS21546 (now being developed by Roche) entered Phase 2 trials for idiopathic pulmonary fibrosis. These represent proof-of-concept that AI can design molecules with genuine therapeutic potential.

Limitations: AI models are constrained by training data quality. Predictions are most accurate for targets with substantial historical bioactivity data. For novel targets or rare diseases with limited data, AI predictions become less reliable. Regulatory agencies remain cautious about AI-designed drugs lacking mechanistic understanding—transparency and interpretability remain challenges.

8. CRISPR-Based Drug Target Identification

Principle: Use CRISPR gene editing to systematically perturb genes in disease-relevant cells, identifying which perturbations produce therapeutic phenotypes. These genes become prioritized drug targets.

Process: Researchers create CRISPR libraries targeting thousands of genes. These are introduced into disease-relevant cell models (e.g., cancer cells, neurons, immune cells). Cells are then challenged with disease stimuli or toxins. Cells with certain gene disruptions survive or recover preferentially—these "winning" genes become candidate drug targets. Compounds can then be screened to modulate these targets.

Advantages: CRISPR screens are unbiased—they don't require researchers to hypothesize which genes matter. They can discover unexpected targets that might not emerge through traditional approaches. They're relatively rapid and increasingly cost-effective.

Real-World Example: Companies like Recursion and others have used genome-scale CRISPR screens to identify targets in neurodegeneration, cancer, and immunology. These identified targets then feed into traditional and AI-powered drug discovery pipelines.

Clinical Translation: CRISPR screens often identify targets, but developing inhibitors against these targets still requires the full drug discovery pipeline—screening, optimization, safety assessment, and clinical testing.

9. DNA-Encoded Compound Libraries (DEL)

Principle: Create millions or billions of chemical compounds, each with an attached DNA barcode. Screen these libraries against targets; winning compounds are identified by sequencing their DNA tags.

Process: In a DEL, each compound carries a unique DNA sequence identifying its structure. Libraries containing up to 10^15 unique compounds are possible. The entire library is exposed to a target protein (or cell-based assay). Compounds that bind the target are captured (e.g., via biotin-streptavidin or immobilization). The DNA barcodes of binding compounds are then sequenced, revealing which structures bound the target.

Advantages: DELs enable screening of extraordinarily large chemical spaces far exceeding what's possible with traditional HTS or even some computational approaches. They can identify unexpected chemical scaffolds that bind intractable targets.

Key Companies:

• X-Chem: Founded in 2000, X-Chem pioneered DEL technology. Several X-Chem-derived compounds have entered clinical development.
• HitGen: A Chinese company specializing in DEL screening services. They've screened libraries against protein-protein interaction targets and kinases, identifying novel lead compounds.
• Nuevolution (acquired by Amgen in 2020): Desarrolló DEL technology for complex multi-component libraries. Amgen acquired the company specifically to integrate DEL into their discovery pipeline.

Clinical Impact: DEL-derived compounds are beginning to enter clinical pipelines, particularly for difficult-to-target protein-protein interactions and protein degradation.

10. Fragment-Based Drug Discovery (FBDD)

Principle: Screen small molecular fragments (MW <300) against targets, identify fragments that bind, then systematically grow them into larger, more potent compounds.

Process: Fragment libraries (typically 1,000-10,000 compounds) are screened against targets using sensitive biophysical methods (surface plasmon resonance, nuclear magnetic resonance, thermal shift assays). Binding fragments are identified, then grown through structure-based design and medicinal chemistry to increase potency and selectivity. The final drug-like molecule often retains the core fragment binding motif while acquiring additional groups improving pharmacology and pharmacokinetics.

Advantages: FBDD can identify novel chemical scaffolds by working in lower molecular weight space. It's particularly effective for targets lacking traditional inhibitors. Many FBDD-derived compounds have better druggability and fewer off-target effects than some HTS-derived leads.

Clinical Example: Several marketed drugs originated from FBDD, including vemurafenib (Zelboraf) for melanoma, venetoclax (Venclexta) for chronic lymphocytic leukemia, and others. Structural Genomics Consortium (SGC) has championed FBDD for difficult targets.

11. Phenotypic Screening Revival

Principle: Screen compounds in complex cellular or organism-level assays measuring disease-relevant phenotypes (e.g., neuronal degeneration, behavioral changes, immune activation) without necessarily understanding the molecular target.

Context: In the 1990s and 2000s, the field shifted heavily toward target-based approaches. But many target-based programs failed in clinical trials—compounds that worked perfectly against their targets didn't improve patient outcomes. This led to recognition that disease complexity often exceeds single-target effects.

Modern Phenotypic Screening: Contemporary approaches combine phenotypic measurement with modern genomics. After identifying a compound producing a desired phenotype in cells or organisms, researchers use CRISPR, transcriptomics, or proteomics to identify the targets responsible for the phenotype. This combines the strengths of phenotypic screening (discovering compounds that genuinely affect disease biology) with mechanistic insight.

Example: Recursion Pharmaceuticals prominently uses phenotypic screening combined with image analysis. Their platform applies high-throughput microscopy to identify compounds affecting disease-relevant cell phenotypes in neurodegeneration, epilepsy, and other conditions. The phenotype drives prioritization; target identification follows.

Advantages: Phenotypic screening can identify compounds affecting complex, multifactorial disease processes. It's less susceptible to false leads from target-based approaches that don't translate to human benefit.

From Discovery to Clinics: The Valley of Death and Eroom's Law

Understanding modern drug discovery requires confronting a sobering reality: the vast majority of compounds that appear promising in the laboratory never reach patients. The journey from initial hit compound to marketed drug is perilous, expensive, and marked by systematic attrition.

The Numbers: Success Rates and Timelines

On average, bringing a single new drug to market requires:

• 10-15 years of development (discovery through regulatory approval)
• $2.6 billion total investment (discovery, preclinical, clinical, regulatory)
• Screening of 5,000-10,000 initial compounds to identify one clinical candidate
• Only 12% of drugs entering Phase 1 clinical trials ultimately receive FDA approval
• 50% failure rate in Phase 2 trials (efficacy failures)
• 60% failure rate in Phase 3 trials (efficacy and safety issues)

These statistics illustrate why pharmaceutical companies prioritize: they must choose targets and compounds with high likelihood of clinical success to justify the enormous investment and lengthy timeline.

The Valley of Death

The "valley of death" refers to the gap between promising laboratory discoveries and successful clinical translation. A compound may show perfect selectivity for a target, bind with nanomolar affinity, and produce dramatic effects in cell-based or animal models—yet fail catastrophically in humans.

Common Causes of Failure:

• Lack of Target Engagement: The compound doesn't reach therapeutic concentrations in the target tissue (brain, specific organ)
• Off-Target Toxicity: The compound binds unintended targets, causing toxicity unrelated to efficacy
• Pharmacokinetic Failure: Poor absorption, rapid metabolism, or rapid elimination prevent therapeutic exposure
• Efficacy Gap: Animal models don't accurately predict human disease. A compound effective in a mouse disease model may not translate to human efficacy
• Regulatory Concerns: Safety signals emerge in clinical trials, or efficacy doesn't meet regulatory thresholds
• Disease Biology Misunderstanding: The target, which seemed disease-relevant based on genetics or prior studies, contributes minimally to disease pathology

Eroom's Law: The Paradox of Declining Productivity

Eroom's Law is Moore's Law spelled backward. Moore's Law posits that computing power doubles every 18-24 months—a phenomenon driving exponential technological progress. Eroom's Law describes the inverse trend in pharmaceutical innovation: despite exponential increases in research spending and technological capability, the number of new drugs approved per research dollar has halved every 9 years since the 1950s.

Historical Context:

• In the 1950s, pharmaceutical companies discovering a single new molecular entity required approximately $100-200 million (inflation-adjusted)
• By 2020, the same achievement required $2.6 billion—more than a tenfold increase in real terms
• This decline occurred despite computers becoming infinitely more powerful, high-throughput technologies enabling screening of millions of compounds annually, and vast expansion of disease-relevant knowledge through genomics and structural biology

Proposed Explanations:

• Low-Hanging Fruit Depletion: The easiest targets (obvious disease mechanisms) have been discovered. Remaining targets are harder to drug (e.g., protein-protein interactions)
• Regulatory Tightening: Standards for safety and efficacy have increased. Drugs that would have been approved in the 1970s would fail modern regulatory review
• Target Complexity: As understanding of disease biology improves, it becomes apparent that disease results from multiple targets and complex interactions. Single-target drugs often prove insufficient
• Loss of Incentive Diversity: Economic consolidation of the pharmaceutical industry has reduced the number of independent research programs pursuing different strategies
• Productivity Measurement Artifacts: Companies increasingly focus on "blockbuster" drugs (billion-dollar annual sales), deprioritizing smaller therapeutic areas. This skews productivity metrics

The Future: AI, Convergence, and Accelerated Discovery

Despite Eroom's Law, the future of drug discovery appears promising. Emerging technologies promise to address specific bottlenecks:

AI and Machine Learning Convergence

Rather than replacing traditional methods, AI appears most powerful when integrated with existing approaches. Combining phenotypic screening with AI target identification, classical pharmacology with computational lead optimization, and CRISPR target discovery with ML-guided compound design creates synergistic workflows. Several companies report that AI-integrated discovery platforms reduce lead optimization timelines by 30-50%.

Biomarker-Driven Patient Selection

Rather than seeking drugs effective in broad patient populations, emerging approaches select patients likely to respond based on biomarkers (genetic, proteomic, imaging). This increases efficacy signals and reduces failed trials. Regulatory agencies increasingly accept biomarker-stratified trials.

Organ-on-Chip and Advanced Models

Traditional animal models poorly predict human drug responses. Organ-on-chip technologies—microfluidic systems recreating human tissue architecture—promise more accurate prediction of toxicity and efficacy. Companies like Emulate and Hesperos are developing these; regulatory agencies are considering their use for safety assessment.

Convergence with Biotechnology

Traditional small-molecule drugs increasingly merge with biologics (antibodies, proteins, cell therapies). Hybrid approaches combining small molecules with targeted delivery, combination therapies addressing multiple targets, and personalized approaches increase success rates.

Further Reading & References

1. Schou M. Lithium in psychiatric therapy and prophylaxis. J Psychiatr Res. 1970;6(67):67-95. [The seminal clinical validation of lithium]
2. Cade JF. Lithium salts in the treatment of psychotic excitement. Med J Aust. 1949;2(10):349-352. [Cade's original guinea pig discovery]
3. Laborit H, Huguenard P. L'hybernation artificielle par moyens pharmacodynamiques et physiques. Presse Med. 1951;59(24):419. [Early chlorpromazine observations]
4. Meanwell NA. Improving drug candidates by design: a focus on physicochemical properties. Chem Res Toxicol. 2011;24(9):1420-1456. [Comprehensive review of drug-likeness principles]
5. Paul SM, Mytelka DS, Dunwiddie CT, et al. How to improve R&D productivity: the pharmaceutical industry's grand challenge. Nat Rev Drug Discov. 2010;9(3):203-214. [Early analysis of Eroom's Law]
6. Scannell JW, Blanckley A, Boldon H, Warrington B. Diagnosing the decline in pharmaceutical R&D efficiency. Nat Rev Drug Discov. 2012;11(3):191-200. [Comprehensive Eroom's Law analysis]
7. Mullard A. Structural biology: why hasn't it revolutionized drug discovery? Nat Rev Drug Discov. 2021;20(2):88-91. [Discussion of structural biology's limited translation]
8. Ratti M, Lampropoulou A, Sivolapenko GB. Artificial intelligence applications in pharmaceutical sciences. Adv Drug Deliv Rev. 2024;205:114979. [Recent AI in drug discovery review]
9. Pye CR, Bertin MJ, Lokey RS, Gerwick WH, Linington RG. Retrospective analysis of natural products provides insights for future antifungal discovery. Proc Natl Acad Sci USA. 2017;114(12):3141-3146. [Natural products and drug discovery]
10. Erlanson DA, Fesik SW, Hubbard RE, Jahnke W, Jhoti H. Twenty years on: the impact of fragments on drug discovery. Nat Rev Drug Discov. 2016;15(9):605-619. [Fragment-based drug discovery review]
11. Hann MM, Leach AR, Harper G. Molecular complexity and its impact on the probability of finding leads for drug discovery. J Chem Inf Comput Sci. 2001;41(3):856-864. [Lipinski Rule of Five context]
12. Fishman MC, Porter JA. Pharmaceuticals: a new grammar for drug discovery. Nature. 2005;437(7060):491-493. [Philosophical perspective on discovery methods]
13. Vamathevan K, Clark D, Soares P, et al. Applications of machine learning in drug discovery and development. Nat Rev Drug Discov. 2019;18(6):463-477. [Comprehensive ML applications review]
14. Fleming A. On the antibacterial action of cultures of a penicillium, with special reference to their use in the isolation of H. influenzae. Br J Exp Pathol. 1929;10(3):226-236. [Fleming's original penicillin paper]
15. Harrison RK. Phase II and phase III failures: 2013–2015. Nat Rev Drug Discov. 2016;15(12):817-818. [Clinical trial failure rate analysis]