What Makes a Molecule Druglike in Drug Design?

Druglike describes a set of physical and chemical properties that make a molecule likely to work as an oral medicine in the human body. The concept emerged from a practical problem: of the millions of compounds a chemist could build, most will never survive the journey from the gut into the bloodstream and then to the tissue where they are needed. Druglikeness is the field’s shorthand for the cluster of traits that separate molecules with a realistic shot at becoming pills from those that are doomed to fail. The idea sounds straightforward, but the rules keep getting rewritten as drug targets get larger, computational tools get sharper, and new drug formats push past the old boundaries.

Where the Idea Came From

In the late 1990s, researchers at Pfizer noticed that many promising-looking compounds were failing in early clinical development because the human body simply could not absorb them. Christopher Lipinski and colleagues analyzed compounds that had successfully moved from Phase I into Phase II clinical studies, correlating the molecules’ computed physical properties with their solubility, permeability, and oral bioavailability. The result was a set of thresholds that became known as the Rule of Five, named because each cutoff is a multiple of five: a molecular weight no greater than 500, a measure of greasiness (the partition coefficient, cLogP) no greater than 5, no more than 5 groups that donate hydrogen bonds, and no more than 10 groups that accept them.¹ The Rule of Five was never meant to be a law. It was a quick mental checklist for chemists: if your molecule violates two or more of these limits, it probably has poor solubility or permeability, and you should think hard before investing more time in it.

What Each Property Actually Controls

To become a working oral drug, a molecule has to do two things that are in tension with each other. It needs enough water-friendliness to dissolve in the gut and travel through the blood, yet enough greasiness to slip across the oily cell membranes that line the intestinal wall and every target tissue. The Rule of Five properties are proxies for this balancing act.

Molecular weight matters because larger molecules have a harder time squeezing through cell membranes. Lipophilicity (cLogP) captures how greasy a compound is; too greasy and it will not dissolve in water, too water-loving and it cannot cross membranes. Hydrogen-bond donors and acceptors are particularly important because every hydrogen bond a molecule can form with water is a bond it must break to enter the fatty interior of a membrane, creating an energy penalty that slows absorption.² One study found that hydrogen-bond donor count was more predictive of human intestinal permeability than acceptor count, and that combining a donor descriptor with a lipophilicity descriptor enabled reasonable predictions of how well compounds would be absorbed.³

Both the volume of a molecule and its hydrogen-bonding potential contribute to how easily it crosses membranes, but the relationship is not a simple linear one. The rate-limiting step depends on the specific molecule: for some compounds, size is the bottleneck; for others, the penalty from breaking hydrogen bonds dominates.⁴

Flexibility and Surface Area

After the Rule of Five caught on, researchers identified additional properties that matter for oral absorption. Two of the most cited are the number of rotatable bonds, which measures how floppy a molecule is, and polar surface area, which captures how much of the molecule’s exterior is covered in oxygen and nitrogen atoms that interact with water. A widely cited study proposed that compounds with ten or fewer rotatable bonds and a polar surface area of 140 square angstroms or less had a much better chance of being orally bioavailable in rats. Reduced polar surface area correlated better with increased permeation rate than lipophilicity did, and a rising rotatable bond count had a negative effect on permeation.⁵

These thresholds, sometimes called the Veber rules, became a standard companion to Lipinski’s original four. But subsequent work showed that the correlations depend on the calculation method used and on the therapeutic class of the compounds being studied, underscoring that these guidelines are statistical tendencies, not hard cutoffs.⁶

When the Rules Get Broken on Purpose

For a generation of medicinal chemists, keeping a molecule within Rule of Five space was almost a reflex. But biology does not care about rules of thumb. Some of the most exciting drug targets today, particularly in oncology and immunology, involve protein-protein interactions that require much larger molecules to disrupt. PROTACs (proteolysis targeting chimeras) are a prime example. These are bifunctional molecules with two active ends joined by a linker, and they typically weigh 700 to 1,000 daltons, well above the 500 threshold. By the old logic, none of them should be orally bioavailable. Yet some are.

Researchers studying PROTAC absorption found that the property limits for these bigger molecules are genuinely wider than the Rule of Five predicts, but not unlimited. Oral absorption was achievable up to a molecular weight of about 950, with up to 14 rotatable bonds and a polar surface area up to 200 square angstroms. The hydrogen-bond donor limit was more nuanced: the apparent ceiling was 4 donors, but nearly all of the well-absorbed compounds with 3 or 4 donors had one or two of those donors satisfied internally, folded back onto the molecule itself. In practice, the effective limit was 2 unsatisfied (solvent-exposed) hydrogen-bond donors.⁷ A separate NMR-based study of PROTAC structures confirmed this upper limit of two exposed donors for oral bioavailability, while showing greater tolerance for other properties like polarity and molecular weight.⁸

The concept behind this is sometimes called chameleonicity: large molecules that can tuck their polar groups inward in a greasy membrane environment and expose them again in water. This conformational trick lets molecules that look too polar on paper behave as if they are far less polar when it counts. It is a major area of current research, and it suggests that druglikeness for the next generation of medicines will be defined less by static property thresholds and more by a molecule’s ability to change its apparent character depending on its surroundings.

Brain Drugs Play by Tighter Rules

If getting a drug across the intestinal wall is hard, getting one into the brain is harder. The blood-brain barrier is lined with tightly packed cells and studded with efflux pumps that actively push foreign molecules back out. Successful central nervous system drugs tend to have a narrower range of molecular weight, lipophilicity, and hydrogen-bond donor and acceptor counts than drugs aimed at other parts of the body.⁹ In practice, CNS drugs are smaller, less polar, and more lipophilic than the average oral drug, but not too lipophilic, because excessively greasy molecules get chewed up by liver enzymes and tend to bind nonspecifically to proteins throughout the body.

To navigate this narrow corridor, drug designers developed multiparameter optimization (MPO) scoring tools tailored specifically to the brain. One widely used CNS MPO tool helped increase the proportion of drug candidates that successfully crossed the blood-brain barrier while also sitting in lower-risk safety space, characterized by lower lipophilicity and higher polar surface area than older CNS compounds.¹⁰ Blood-brain barrier permeability adds an extra constraint on top of all the usual druglikeness requirements, making CNS drug discovery one of the most difficult areas in the field.¹¹

Solubility Is Often the Real Bottleneck

A compound can have perfect permeability across membranes and still fail if it does not dissolve in the gut. The Biopharmaceutical Classification System (BCS) groups drugs into four classes based on their solubility and permeability, and it has become a standard regulatory tool for deciding when generic drugs can skip certain clinical tests.¹² The catch is that the BCS solubility test is done in plain water at various pH values, which does not reflect what actually happens inside the intestine. The small intestine is not just water: it contains bile salts and lipid aggregates that help dissolve greasy compounds.

One study found that the majority of BCS Class 2 compounds (those labeled “low solubility, high permeability”) were actually fully absorbed after oral dosing. When researchers tested solubility at the pH of the small intestine or in simulated intestinal fluid containing lipids, the complete dose of many supposedly poorly soluble drugs dissolved just fine. For acidic compounds, simply measuring solubility at intestinal pH rather than the standard acidic pH reclassified them correctly. For very lipophilic compounds, the presence of natural intestinal lipids did the job.¹³ This matters for druglikeness assessments because a molecule that looks poorly soluble by the standard criteria might actually dissolve well enough in the conditions that exist inside a living gut.

Lipophilicity and How Quickly the Body Clears a Drug

Getting a drug absorbed is only half the battle. The same liver enzymes that metabolize food and toxins are constantly working to break down drug molecules. A key family of these enzymes, the cytochrome P450 system, has a well-documented preference for greasy substrates. Research has shown that for a given metabolic efficiency, the greasier a compound is, the more avidly it binds to these enzymes and the faster it gets cleared from the body.¹⁴ This creates a paradox: you want enough greasiness for good membrane permeability, but every notch upward on the lipophilicity scale makes the liver more eager to destroy the molecule before it can do its job.

High lipophilicity also brings safety concerns. Greasy molecules tend to bind to unintended targets, including ion channels in the heart. One case study involved a promising kinase inhibitor that hit the hERG ion channel, a well-known cardiac safety liability. By dialing down lipophilicity through careful chemical modifications, the team reduced hERG binding dramatically while maintaining potency against the intended target.¹⁵ Managing lipophilicity is arguably the single most important lever in druglike design: get it right and absorption, metabolism, and safety all tend to fall into line.

Scoring Druglikeness Beyond Simple Cutoffs

The Rule of Five and the Veber rules are pass/fail: a compound either sits inside the limits or it does not. But medicinal chemists have long wanted a way to say how druglike a compound is, not just whether it crosses a line. Several continuous scoring systems have been developed for this purpose. One of the most widely adopted is the quantitative estimate of drug-likeness (QED), which looks at the distributions of multiple properties across known drugs and produces a single score reflecting how closely a candidate molecule matches those distributions. The QED approach is transparent and easy to implement, and it lets researchers rank compounds against each other rather than just sorting them into “pass” and “fail” bins.¹⁶

Machine learning has pushed these approaches further. Algorithms trained on large datasets of known drugs and non-drugs can learn to recognize structural patterns associated with druglikeness, achieving classification accuracies above 90% in some benchmarks.¹⁷ Deep learning techniques that operate directly on molecular graphs, rather than relying on pre-calculated properties, have further improved prediction accuracy by capturing relationships that simple property cutoffs miss.¹⁸ These tools do not replace chemical intuition, but they let researchers screen virtual libraries of millions of compounds overnight and prioritize the most promising candidates for synthesis.

Molecules That Fool You

Druglikeness filters are designed to keep compounds in. There is a parallel set of filters designed to keep bad actors out. Pan-assay interference compounds, or PAINS, are molecular troublemakers that show up as apparent hits in almost every screening assay they are tested in. They generate false positive signals through mechanisms that have nothing to do with genuinely binding to a drug target: they might aggregate in solution, react chemically with assay components, or interfere with fluorescent detection.¹⁹

Electronic filters that recognize common PAINS substructures can flag these compounds in seconds, and they are now in widespread use at pharmaceutical companies and academic screening centers alike.²⁰ The original PAINS filter identified structural fragments that standard reactivity filters missed entirely, catching a class of problematic compounds that could otherwise eat up years of fruitless optimization.²¹ A molecule can look perfectly druglike by every property metric and still be a PAINS compound, which is why these filters operate as a complementary layer rather than a replacement for druglikeness scoring.

Drug Transporters and the Limits of Passive Diffusion

All of the druglikeness rules described so far assume that a molecule gets into cells by passively drifting across membranes. In reality, cells are studded with transport proteins that actively pump molecules in or out. Some transporters help drugs get absorbed by ferrying them into intestinal cells. Others, particularly the efflux pumps, work against you by grabbing drug molecules that have made it into a cell and spitting them back out. The role of these transporters as determinants of drug permeability has become increasingly recognized, and they add a layer of complexity that static property calculations cannot fully capture.²²

For the blood-brain barrier, efflux pumps are a particularly serious concern. A compound can have ideal size, lipophilicity, and hydrogen-bonding profiles and still fail to reach the brain because P-glycoprotein, one of the most studied efflux transporters, pumps it right back out. This is one reason CNS druglikeness guidelines put so much emphasis on keeping polar surface area in a sweet spot: too high and the molecule cannot cross the membrane at all, too low and efflux pumps recognize it more readily as a substrate.

Oral Versus Non-Oral Routes

Druglikeness is most commonly discussed in the context of oral drugs, because swallowing a pill is the default delivery route and because oral absorption imposes the strictest physical-chemical constraints. When researchers examined FDA-approved drugs delivered by non-oral routes, including inhaled, ophthalmic, and transdermal products, over 98% still fell within the Rule of Five limits. But that statistic is misleading: it reflects what has been tried, not what is possible. Further analysis showed that inhaled drugs can work well outside Rule of Five space, including large hydrophilic molecules like proteins, because the lungs offer an enormous, thin surface with direct access to the bloodstream.²³

Transdermal drugs actually face tighter constraints than oral ones: the skin is a far more formidable barrier than the gut lining, though new delivery technologies like microneedle patches are changing this. Oral drugs specifically tend to be lighter and have fewer hydrogen-bond donors, acceptors, and rotatable bonds than drugs given by other routes, with injectable drugs being the most different from oral ones in these properties.²⁴ The takeaway is that druglikeness is route-dependent. A molecule that is a poor oral candidate might be an excellent inhaled or injectable drug.

How Intramolecular Tricks Expand What Counts as Druglike

One of the more fascinating recent insights is that molecules can game the druglikeness system by folding in on themselves. When a hydrogen-bond donor on one part of a molecule finds a hydrogen-bond acceptor on another part of the same molecule, it forms an intramolecular hydrogen bond. This effectively hides the polar group from the surrounding environment. Research on piracetam, a small drug molecule, showed that such an intramolecular bond lowered the energy barrier for crossing a lipid membrane by a significant margin, boosting permeability and partially compensating for the penalty the molecule would otherwise pay for entering the greasy membrane interior.²⁵

This is the same phenomenon driving oral absorption in PROTACs, scaled up. Medicinal chemists are increasingly designing molecules with built-in intramolecular interactions, engineering them to present a low-polarity face in membranes and then open up their polar groups once they reach their aqueous target. The traditional property-based filters count every hydrogen-bond donor the same, whether it is exposed to water or buried inside the molecule. As these chameleonic strategies become better understood, the next generation of druglikeness models will likely incorporate conformational analysis rather than just counting functional groups.

Natural Products and Chemical Space

Plants, fungi, and bacteria have been producing biologically active molecules for hundreds of millions of years, and natural products remain a rich source of drug candidates. Their chemical space, the range of structural and property diversity they cover, is broader than that of typical synthetic compound libraries. Natural products tend to include more complex ring systems, more oxygen atoms, and more chiral centers than what comes out of a standard medicinal chemistry lab. Some of the most important drugs in history (penicillin, taxol, cyclosporine) are natural products or derived from them, and many of these violate Rule of Five guidelines spectacularly.

Cyclosporine, the immunosuppressant, has a molecular weight over 1,200 and a dozen hydrogen-bond donors. It is orally bioavailable because its backbone can form intramolecular hydrogen bonds in lipid environments, the same chameleonic trick now being reverse-engineered for PROTACs. Natural products serve as a reminder that biological evolution solved the oral absorption problem long before chemists wrote down rules about it, and that the apparent boundaries of druglikeness may say more about the limits of our synthetic methods than about the limits of biology.