Drug discovery is the process of finding new molecules that could treat or cure a disease. It covers everything from identifying a biological cause of illness to designing a chemical compound that can safely interfere with that cause. On average, moving a new drug from initial concept to approved therapy takes about 10 years and costs over $2.5 billion, with roughly 90% of candidates failing somewhere along the way. The discovery phase itself is just the front end of that journey, but it’s where the most critical scientific decisions are made.
How Drug Discovery Differs From Drug Development
The terms “drug discovery” and “drug development” are sometimes used interchangeably, but they refer to different stages. Discovery is the creative, exploratory phase: scientists figure out what’s going wrong in a disease, find a biological target to intervene on, and design or screen for molecules that hit that target. Development is everything that comes after, including large-scale clinical trials in humans, manufacturing scale-up, and regulatory review.
The FDA outlines the full pipeline in five steps: discovery and development, preclinical research, clinical research, FDA review, and post-market safety monitoring. Drug discovery feeds into the first two of those steps. It ends when researchers have a promising compound with enough safety data to file an Investigational New Drug (IND) application, which is the formal request to begin testing in people.
Finding the Right Target
Every drug discovery effort starts with a question: what exactly is going wrong in the body, and where can a drug intervene? The answer is called a “target,” usually a specific protein, enzyme, or receptor that plays a clear role in the disease process. Researchers validate a potential target using three major types of human data: whether the protein is expressed in the relevant tissue, whether genetics link it to the disease, and whether clinical experience supports the connection.
In Alzheimer’s research, for example, scientists have spent years evaluating a protein fragment called beta-amyloid as a drug target. Abnormal deposits of this protein are a hallmark of the disease. Researchers at one point traced the signaling pathway that produces the harmful form of beta-amyloid, then showed they could block its production by targeting a specific receptor in nerve cells. That kind of detective work, mapping a disease mechanism down to a single molecular interaction, is what target identification looks like in practice. It can take years before scientists are confident enough in a target to start designing drugs against it.
Screening for Compounds
Once a target is validated, the next step is finding a molecule that interacts with it. The traditional approach is high-throughput screening, where automated systems test hundreds of thousands of chemical compounds against the target to see which ones have any effect. This brute-force method typically yields only a handful of weak “hits” after months of work.
A hit is not a drug. It’s simply a molecule that shows some activity against the target. Researchers then refine these hits into “lead” compounds by tweaking their chemical structure to improve potency, selectivity, and basic safety properties. A major part of this refinement involves testing how well the compound is absorbed, distributed through the body, metabolized, and eliminated. These characteristics, collectively known as ADME properties, determine whether a molecule that works in a test tube could ever work in a living person. Scientists increasingly use human or “humanized” tissues in lab-based screens to get more reliable predictions earlier in the process.
Designing Drugs From 3D Structures
Not all drug discovery relies on screening massive libraries of compounds. A more targeted approach called structure-based drug design uses the three-dimensional shape of a target protein to build a molecule from scratch. If researchers can map the binding cavity of a protein (the pocket where a drug would need to attach), they can design a molecule that fits into it like a key in a lock.
There are two main techniques for this. In fragment linking, scientists identify small chemical fragments that interact with different parts of the binding pocket, then connect them into a single complete molecule. In fragment growing, they start with one small piece inside the pocket and build outward, adding atoms or chemical groups guided by the shape and chemistry of the cavity. Both approaches require high-quality structural data about the target protein, typically obtained through X-ray crystallography or similar imaging methods. This rational design approach can be faster and more efficient than blind screening because it eliminates millions of unlikely candidates from the start.
Preclinical Testing
Before any compound reaches a human volunteer, it must pass through preclinical testing to establish that it isn’t dangerously toxic. This involves two types of studies: in vitro testing (experiments on cells or tissues in a lab) and in vivo testing (experiments in living animals). The FDA requires these studies to follow Good Laboratory Practices, a set of regulations covering how studies are conducted, who performs them, what equipment is used, and how results are reported.
Preclinical studies are typically small, but they must produce detailed information on dosing and toxicity levels. Researchers need to determine the dose range where the drug is effective, the dose where it becomes harmful, and what organs or systems are most affected by toxicity. This data becomes part of the IND application, alongside chemistry and manufacturing details, a summary of all known pharmacological effects, and a protocol for the first planned human study. Only after the FDA reviews this package and raises no objections can the compound move into clinical trials.
Drug Repurposing as a Shortcut
Not all drug discovery starts from zero. Drug repurposing (also called repositioning) takes medications that are already approved for one condition and investigates whether they work for a different one. Because these drugs have already been through safety testing in humans, the timeline and cost can drop dramatically.
Repurposing gained visibility during the COVID-19 pandemic, when researchers screened databases of approved drugs for antiviral activity against SARS-CoV-2. From one such screening effort, 35 candidates were flagged for further study, and 10 reached advanced clinical trials. Computational methods, including structure-based analysis and deep learning, have made repurposing increasingly systematic. Rather than stumbling onto a new use by accident, researchers can now model how an existing drug’s shape and chemical properties might interact with targets in a completely different disease. This strategy has proven especially active in rare and hard-to-treat diseases, where the economics of building a drug from scratch are particularly challenging.
How AI Is Changing the Process
Artificial intelligence is reshaping nearly every stage of drug discovery. Machine learning models trained on large datasets of known compounds can predict whether a new molecule will be effective against a target, how toxic it might be, and how well the body will absorb it. These predictions used to require months of lab work for each compound. AI can generate them in hours for thousands of candidates simultaneously.
One of the most promising applications is de novo drug design, where deep learning algorithms propose entirely new molecules with specific desired properties like solubility, potency, and low toxicity. A model trained on the structures and properties of known drugs can generate novel chemical structures that have never been synthesized before, then rank them by likelihood of success. This doesn’t replace lab validation, but it dramatically narrows the field. Instead of screening hundreds of thousands of compounds to find a few weak leads, researchers can start with a shortlist of computationally optimized candidates. The World Economic Forum has noted that these AI-driven approaches are beginning to compress timelines that historically stretched over a decade, though the full impact on approval rates is still playing out.
Why Most Candidates Fail
The 90% failure rate in drug discovery and development isn’t a sign of bad science. It reflects the enormous biological complexity of treating disease in a living human body. A molecule might bind perfectly to its target in a test tube but get broken down by the liver before reaching its destination. It might work in animal models but trigger an immune reaction in people. It might reduce a biomarker in clinical trials without actually improving symptoms.
Each stage of discovery is designed to eliminate these failures as early and cheaply as possible. Target validation weeds out biological dead ends. ADME screening catches molecules with poor pharmacokinetics. Preclinical toxicity testing flags safety risks before anyone is exposed. The entire pipeline is essentially a series of filters, each one more expensive and time-consuming than the last, designed to ensure that only the most promising compounds move forward. The ones that make it through represent a tiny fraction of what started in the lab, but they arrive at clinical trials with the strongest possible evidence that they might actually work.