Target Validation and Lead Optimization: Reducing Attrition Before the Clinic

Most drug programs do not fail because the chemistry was poor. They fail because the target was wrong, or because nobody tested the biology hard enough before money started to flow. Target validation and lead optimization are where those risks are cheapest to find. This guide walks through the discovery funnel, what the attrition data tell us, and how sponsors can build stronger evidence before they nominate a candidate.

Why Early Discovery Decides Clinical Outcomes

The numbers are sobering. BIO, Informa and QLS Advisors tracked clinical programs from 2011 to 2020 and found an overall likelihood of approval (LOA) from Phase I of 7.9%. Phase II had the lowest transition rate of any stage, at 28.9%. The report calls it the largest hurdle in drug development (BIO, 2021).

Phase II is usually the first real test of whether hitting the target changes the disease. When it does not, the problem often started years earlier. A Tufts CSDD analysis of drugs entering human testing from 2000 to 2009 found that efficacy drove 54% of Phase II failures and 52% of Phase III failures (Applied Clinical Trials, 2013). Duxin Sun and colleagues put lack of clinical efficacy at 40–50% of failures, with unmanageable toxicity at about 30% and poor pharmacokinetic properties at 10–15% (Sun, 2022).

The message is clear. Better targets and better-characterized molecules are the main levers a sponsor controls.

The Discovery Funnel at a Glance

Early discovery usually runs through a familiar sequence. Hughes and colleagues describe it as target identification, target validation, assay development and screening, hit identification, and lead optimization, ending with selection of a candidate for clinical development. The full path from idea to launch can take 12–15 years and cost more than $1 billion (Hughes et al., 2011).

Stage	Key Question	Typical Output
Target ID	Is this target linked to the disease?	Target hypothesis
Target validation	Does modulating it change disease biology?	Validation package
Hit identification	Can we find molecules that engage it?	Confirmed hits
Hit-to-lead	Is there a tractable series?	Lead series with SAR
Lead optimization	Can one molecule meet the full profile?	Development candidate

Building Strong Target Validation Evidence

No single experiment validates a target. Strong packages layer several kinds of evidence, each answering a different question about causality, tractability and safety.

Human Genetics

Human genetic evidence is the most powerful predictor we have. Nelson et al. found that the share of drug mechanisms with direct genetic support rose from 2.0% at the preclinical stage to 8.2% among approved drugs, and estimated that selecting genetically supported targets could double clinical success rates (Nelson et al., Nature Genetics, 2015).

Minikel et al. revisited the question with larger datasets. Drug mechanisms with genetic support had about 2.6 times the probability of success. The benefit grew with confidence in the causal gene, was highest for Mendelian (OMIM) evidence at 3.7, and was most pronounced in Phases II and III. Yet only 4.8% of target–indication pairs active in Phases I to III had germline genetic support (Minikel et al., Nature, 2024).

“A target with no human evidence behind it is a hypothesis, not an asset. Treat it that way in your budget.”

Functional and Translational Tools

• CRISPR knockout and activation screens. Pooled or arrayed screens test which genes drive a disease phenotype, and knockouts confirm that loss of the target reproduces the expected effect.
• Tool compounds and antibodies. Selective probes show whether pharmacological modulation, not just genetic loss, delivers the effect.
• Disease models. Patient-derived cells, organoids and animal models test the biology in more relevant systems. Their limits should be stated up front.
• Biomarkers. Target engagement and pathway markers let you prove the mechanism in early clinical studies, and patient selection markers improve the odds later.

Biomarkers deserve special attention. In the BIO dataset, programs using patient preselection biomarkers had an LOA from Phase I of 15.9%, about double the 7.6% for programs without them. The biggest gap came at Phase II: 46.3% versus 28.3% (BIO, 2021).

From Hit to Lead to Candidate

Once a target is validated and screening produces hits, chemistry teams triage them. Hit-to-lead work confirms activity, removes assay artefacts, and checks that a series has a clear structure–activity relationship (SAR) and room to improve. A good lead series is tractable, patentable and free of obvious red flags.

Lead optimization then refines the series into a single candidate. The core parameters are well known:

• Potency. Activity at the target and in cells, ideally linked to a predicted human dose.
• Selectivity. Clean profiles against related targets and standard off-target panels.
• DMPK and ADME. Solubility, permeability, metabolic stability, plasma protein binding and drug–drug interaction risk.
• Safety. Early in vitro flags such as hERG, cytotoxicity and genotoxicity, followed by exploratory in vivo tolerability.
• Developability. Solid form, stability and a synthetic route that can scale.

These properties often pull against each other. Raising potency can raise lipophilicity, which can worsen solubility and metabolic stability. Multiparameter optimization (MPO) tools address this by scoring each compound against a weighted set of property targets, so teams compare overall quality rather than chasing one number.

Where AI and Outsourcing Fit

AI and machine learning are now routine in target prioritization, virtual screening, property prediction and de novo design. Early results are encouraging but should be read carefully. A BCG analysis of AI-native biotech pipelines, published in Drug Discovery Today, reported Phase I success rates of 80–90% for AI-discovered molecules, against 40–65% historically. Phase II success was about 40%, in line with industry averages, and based on a limited sample (HTWorld, 2024; BCG, 2024).

In other words, AI appears to help make molecules with good drug-like properties. It has not yet shown that it solves the efficacy problem, which is still a question of target biology.

Outsourcing is another lever. Many small sponsors run discovery through contract research organizations (CROs) for chemistry, screening, DMPK and in vivo pharmacology. This gives access to capacity and specialist skills without fixed overhead. It also means the sponsor must own the scientific strategy, set clear acceptance criteria, and keep data in a form that will stand up to diligence. Sponsors building cell-based assays should also think early about cell bank quality; our guides to cell line development and mycoplasma testing cover the basics.

Decision Gates and Candidate Selection

Structured governance makes a measurable difference. AstraZeneca introduced its 5R framework, which asks whether a program has the right target, right tissue, right safety, right patients and right commercial potential. Success rates from candidate nomination to Phase III completion rose from 4% in 2005–2010 to 19% in 2012–2016 (Morgan et al., Nature Reviews Drug Discovery, 2018).

What This Means for Biotech Leaders and Investors

For founders, validation evidence is the asset. A program with genetic support, a clear engagement biomarker and a well-documented lead is easier to finance, partner and license. For investors, the quality of the validation package is a better guide to risk than the elegance of the chemistry.

Three practical steps help. Write a target candidate profile before lead optimization starts. Invest in human genetic and biomarker evidence early, when it is cheap relative to a failed Phase II. And build decision gates with explicit stop criteria, whether the work is in-house or at a CRO.

Terraforme Biosciences helps sponsors assess target evidence, shape development strategy, select and manage discovery partners, and prepare validation packages for investors and licensing partners. Learn more about our strategy and business development services.

Conclusion

Attrition is not random. Most clinical failures trace back to weak target biology or molecules that were not optimized against the full profile. Human genetics, biomarkers and disciplined lead optimization will not guarantee success, but the data show they shift the odds in a meaningful way. The cheapest place to fail is before the first patient is dosed.