Loading field map
Drug discovery is the systematic process of identifying and optimizing chemical compounds or biological molecules that can modulate a disease-related biological target, ultimately producing a candidate medicine suitable for clinical testing. It is the early, research-intensive phase of pharmaceutical development, distinct from the later stages of clinical trials, regulatory review, and manufacturing scale-up. The field sits at the intersection of biology, chemistry, and medicine, and its central challenge is deceptively simple: among millions of possible molecules, find the few that are both effective against a disease and safe enough for humans, then prove it well enough to justify the enormous cost of testing in people.
The fundamental difficulty of drug discovery is the vastness of chemical space—the set of all possible small molecules—combined with the biological complexity of disease. Estimates of the number of drug-like molecules that could theoretically be synthesized run into the billions or more, yet only a tiny fraction will interact usefully with any given biological target. A target is typically a protein whose activity contributes to a disease, such as an enzyme, receptor, or signaling molecule. The goal is to find a molecule that binds to that target and modulates its activity in a beneficial way, while avoiding interactions with other proteins that could cause toxicity.
The stakes are extraordinarily high. Developing a new medicine takes more than a decade and costs in the billions of dollars when accounting for failures. The attrition rate is severe: the vast majority of compounds that enter preclinical development never reach patients, and most drugs that enter human trials fail, often because they are unsafe or simply do not work in people as they did in laboratory models. This economic and scientific pressure shapes every decision in the field, from which targets to pursue to which chemical series to advance.
Drug discovery has roots in traditional medicine, where natural products—plant extracts, minerals, and animal parts—were used empirically for centuries. The modern field emerged in the late nineteenth and early twentieth centuries when scientists began to isolate the active ingredients in these remedies and to synthesize new compounds based on chemical knowledge. The German physician Paul Ehrlich coined the term "magic bullet" to describe his vision of a chemical that could selectively kill a pathogen without harming the host, and his work on arsenic compounds for syphilis in the early 1900s is often cited as a foundational example of systematic screening.
For much of the twentieth century, drug discovery was dominated by what might be called the empirical screening approach. Pharmaceutical companies maintained large libraries of chemical compounds and tested them against whole organisms or tissue preparations, looking for any sign of biological activity. This approach produced many important drugs, including antibiotics, antihistamines, and anti-inflammatory agents, but it was largely a matter of trial and error. The mechanism of action of a successful compound might be unknown, and optimization relied on medicinal chemists making structural modifications and testing the results.
A major shift began in the 1970s and 1980s with the rise of molecular biology and the concept of rational drug design. As scientists identified specific proteins involved in disease pathways, it became possible to design molecules that would fit into a target's active site or binding pocket, guided by knowledge of its three-dimensional structure. This approach was enabled by techniques such as X-ray crystallography, which could reveal the atomic structure of proteins, and by computational chemistry, which could model how a drug candidate might interact with its target.
The 1990s brought another transformation with the genomics revolution. The sequencing of the human genome and the development of high-throughput screening—automated systems that could test hundreds of thousands of compounds against a target in a matter of days—expanded the scale of discovery dramatically. For the first time, researchers could systematically search large chemical libraries for hits against specific molecular targets implicated in disease.
Contemporary drug discovery is not a single method but a portfolio of complementary approaches, each with its own strengths, limitations, and historical trajectory. These approaches are best understood not as rival schools that displaced one another but as tools that are often combined in practice.
The dominant paradigm since the 1990s, target-based discovery begins with a hypothesis: that a particular protein plays a causal role in a disease and that modulating its activity will produce a therapeutic benefit. Researchers express the target protein, develop an assay to measure its activity, and then screen large compound libraries to find molecules that inhibit or activate it. Hits are validated, optimized for potency and selectivity, and then tested in cellular and animal models of the disease.
This approach has a clear logic: it focuses effort on a defined molecular mechanism and enables rational optimization. However, it has a critical weakness. A target that is validated in laboratory models may not be relevant to human disease, and a compound that modulates the target perfectly may still fail because the disease involves redundant pathways or because the target's role in healthy tissues causes side effects. The high failure rate of target-based programs in clinical trials has led to criticism that the approach oversimplifies disease biology.
An older approach that has seen a resurgence, phenotypic screening tests compounds against a disease-relevant cellular or organismal model without prior knowledge of the molecular target. For example, a screen might look for compounds that kill cancer cells or that alter the behavior of neurons in a dish. If a compound shows the desired effect, researchers then work backward to identify its target—a process called target deconvolution.
Phenotypic screening has the advantage of starting with a functional readout that may capture aspects of disease biology that target-based approaches miss. It was responsible for many of the most successful drugs of the twentieth century, including aspirin and many antibiotics. Its disadvantages are that it is often harder to optimize a compound without knowing its target, and that the cellular models used may not faithfully represent the human disease. The approach has been revived in recent years partly because of the recognition that target-based discovery, for all its elegance, has not delivered the expected productivity gains.
This is not a separate discovery strategy but a set of techniques that inform both target-based and phenotypic approaches. When the three-dimensional structure of a target protein is known—from X-ray crystallography, cryo-electron microscopy, or computational prediction—researchers can design molecules that fit the binding site, or they can optimize existing hits by visualizing how they interact with the protein. This approach has been particularly successful for enzymes with well-defined active sites, such as HIV protease inhibitors and many kinase inhibitors used in cancer.
The limitation of structure-based design is that it requires a high-resolution structure, which is not always available, and that binding affinity is only one determinant of drug efficacy. A molecule that fits the target perfectly may still be poorly absorbed, rapidly metabolized, or toxic. Structure-based design is therefore best understood as a powerful optimization tool within a broader discovery program rather than a complete methodology.
Computational methods have been part of drug discovery since the 1980s, initially for modeling molecular structures and predicting binding. In recent years, machine learning and artificial intelligence have expanded the role of computation dramatically. Algorithms can now predict protein structures, screen virtual libraries of billions of compounds, predict toxicity and metabolism, and design novel molecules with desired properties.
These methods are promising because they can explore chemical space far more efficiently than physical screening and can integrate diverse data types. However, they are limited by the quality of the data they are trained on and by the fundamental challenge that computational predictions must ultimately be confirmed experimentally. The field is young, and the extent to which AI will transform drug discovery remains an open question. What is clear is that computational tools are increasingly integrated into every stage of the process rather than replacing experimental work.
The approaches described above were developed primarily for small-molecule drugs—traditional chemical compounds with molecular weights under about 900 daltons. A substantial and growing portion of modern drug discovery concerns biologics: large molecules such as antibodies, proteins, and nucleic acids. Biologics are discovered through different processes, often involving immunization of animals to generate antibodies or engineering of proteins to enhance their properties. They offer the advantage of high specificity—an antibody can be designed to bind a single target with remarkable precision—but they are generally more expensive to manufacture, must be injected rather than taken orally, and can provoke immune responses.
The discovery of biologics has its own logic and methods, but it shares the fundamental challenge of small-molecule discovery: identifying a molecule that modulates a disease-relevant target with acceptable safety. Many pharmaceutical companies now run parallel small-molecule and biologic programs against the same target, and the choice between them depends on the nature of the target and the disease.
Contemporary drug discovery is best understood as a staged pipeline, though in practice the stages overlap and loop back on each other. The process typically begins with target identification and validation, in which researchers gather evidence that a protein is causally involved in a disease. This evidence may come from human genetics, from animal models, or from analysis of diseased tissues. The goal is to reduce the risk of pursuing a target that will not matter in patients.
Once a target is chosen, researchers develop an assay—a test that measures the target's activity—and screen for hits. High-throughput screening can test hundreds of thousands of compounds, while fragment-based screening starts with very small molecules and builds them up into larger ones. Hits are confirmed, checked for artifacts, and then enter a lead optimization phase in which medicinal chemists synthesize analogs to improve potency, selectivity, and drug-like properties. This phase is iterative and can take years.
Promising leads are then evaluated in preclinical studies, which include tests in animal models of the disease and comprehensive safety assessments. Only a small fraction of compounds survive this stage to become investigational new drugs, at which point they enter human clinical trials. The transition from discovery to development is a critical handoff: discovery scientists hand their optimized compound to development scientists who will formulate it, scale up its manufacture, and design the clinical trials.
Throughout this pipeline, several cross-cutting disciplines play essential roles. Pharmacology—the study of how drugs interact with biological systems—provides the framework for understanding a compound's absorption, distribution, metabolism, excretion, and toxicity. Toxicology assesses safety. Pharmacokinetics and pharmacodynamics relate drug concentration to effect. And clinical pharmacology bridges the gap between laboratory findings and human responses.
Several features of the current landscape are likely to persist. First, drug discovery remains a high-risk, high-reward enterprise in which most projects fail. The field has learned to manage this risk through portfolio diversification—pursuing multiple targets and multiple compounds—and through increasingly sophisticated methods for failing fast, identifying unpromising compounds early before too much money is spent.
Second, the field is becoming more data-intensive and more interdisciplinary. The integration of genomics, proteomics, and other "omics" data with chemical information and clinical outcomes is creating opportunities for more precise target selection and patient stratification. The hope is that by understanding which patients are most likely to respond to a particular drug, discovery can be more efficient and treatments more effective.
Third, the boundaries of what counts as a drug are expanding. Beyond small molecules and biologics, the field now includes cell and gene therapies, RNA-based medicines, and other modalities that would have been science fiction a few decades ago. These new modalities require new discovery methods and new ways of thinking about what a medicine is.
Finally, the relationship between drug discovery and the broader health-care system is under pressure. The high cost of drugs, the debate over how to price innovation, and the need for drugs that address neglected diseases in low-income countries are all shaping the incentives and priorities of the field. Drug discovery is not purely a scientific endeavor; it is embedded in a complex economic and regulatory environment that determines which diseases are pursued and which treatments reach patients.
The field's enduring challenge is to reconcile the promise of molecular precision with the messiness of human biology. Every new technology—whether high-throughput screening, structural biology, or artificial intelligence—has raised hopes of making drug discovery faster and more predictable. Each has delivered real advances, and each has encountered the same fundamental obstacle: diseases are complex, patients are diverse, and the human body is remarkably good at resisting interventions. The most successful drug discovery programs are those that combine multiple approaches, remain humble about what can be predicted, and maintain the discipline to test every hypothesis rigorously in the laboratory and, ultimately, in patients.