Fragment-to-lead expansion is the phase where ADMET properties start mattering in earnest. Your fragment hit was selected partly for its low molecular weight and simple structure, which typically produce acceptable ADMET predictions by default. As you add bulk to improve potency, that default changes, and different target classes impose very different ADMET requirements on what an acceptable lead actually looks like.
Treating all ADMET properties with equal weight at this stage is a common source of wasted synthesis effort. The fatal properties depend on the target biology, and getting that prioritization wrong means synthesizing compounds that fail the wrong property filter.
Defining the primary ADMET axis by target class
Before ranking ADMET flags in your compound series, establish which property is the program-level constraint for your specific target. This determination should happen before any compound prioritization, not after you get the experimental data back.
For CNS targets: blood-brain barrier penetration is the primary axis. A compound that does not reach the CNS cannot engage the target regardless of its potency. The relevant computed properties are cLogP in the range 1 to 3, molecular weight below 450 Da for oral CNS candidates, topological polar surface area below 90 angstroms squared, and P-glycoprotein efflux ratio. Of these, Pgp efflux is the most frequently missed at the fragment expansion stage: many medicinal chemists optimize lipophilicity and molecular weight correctly but inadvertently introduce Pgp substrates through amide bonds and tertiary amine groups that interact with the efflux transporter. A predicted Pgp efflux ratio above 2 in a CNS program is a hard flag, not a soft one.
For gut-restricted targets, including intestinal kinases, certain GPCR targets expressed primarily in the GI mucosa, and some anti-infective programs targeting gut pathogens: the calculus inverts. You want oral bioavailability that reaches the intestinal mucosa but does not cross systemically. Compounds with high permeability, low Pgp efflux, and moderate metabolic stability in intestinal microsomes are often penalized by a CNS-style ADMET evaluation that scores their permeability as too high.
For oral systemic targets, covering the majority of small molecule programs: hepatic first-pass metabolism dominates. CYP3A4 and CYP2D6 metabolic liability are the primary stability concerns. Compounds with intrinsic clearance above 10 mL/min/kg in human liver microsome predictions will struggle to reach systemic exposure at oral doses that are practical for a drug program.
The Lipinski rule problem in fragment expansion
Ro5 compliance is a useful starting-gate screen, not a sufficient ADMET evaluation for any specific target class. The fragment you started with almost certainly passes Ro5. The problem is that the route to potency often requires adding molecular weight and lipophilicity simultaneously, and the Ro5 cutoffs were derived from a diverse dataset of oral drug candidates, not from the specific physicochemical requirements of any single target biology.
A fragment expansion that stays within Ro5 while optimizing against a CNS target may still fail the Pgp efflux criterion that actually matters for brain penetration. A compound respecting Ro5 that has been extended with a lipophilic tail for back-pocket binding affinity may have acquired a CYP3A4 liability that Ro5 does not flag.
When Alkira reports a tox flag on a compound during fragment expansion, the flag is not solely a Ro5-style check. It reflects predicted liabilities across a broader property set including metabolic stability, hERG channel inhibition, reactive metabolite potential, and protein binding. The weighting of those flags in the composite tox score follows the general ADMET flag hierarchy, but the specific flags most relevant to your program depend on the target class biology we described above.
Metabolic stability: when it matters and when it can wait
In early fragment expansion, predicted metabolic instability is a concern to track but not always a reason to deprioritize. If you are generating SAR data from biochemical assays with purified protein in vitro, the metabolic stability of your compounds is irrelevant to that data generation. The compounds do not need to survive hepatic first-pass to tell you whether the methyl group at position 4 improves binding potency by 3-fold.
Where metabolic stability becomes a prioritization factor is when you are about to decide which compounds to advance to cell-based assays with longer incubation times, where metabolite generation can confound activity data, or when you are selecting compounds for in vivo work. At those decision points, predicted hepatic clearance becomes a primary filter, not a secondary one.
The practical consequence for fragment expansion is that a high metabolic clearance prediction in early ranking is a yellow flag, not a red one. Deprioritizing a structurally interesting compound because of metabolic instability predictions when you have not yet generated enough SAR data to guide optimization toward a better metabolic profile is premature. Keep the compound as a backup. Mark it for metabolic soft-spot analysis once the series is more advanced. Do not cut it from the SAR exploration queue on the basis of a single ADMET prediction when the binding properties are compelling.
hERG: the universal concern that is often misapplied
hERG inhibition prediction is reported in Alkira's tox output as a specific flag because it is one of the most common reasons compounds fail safety screening, and because a minority of well-known structural features, particularly basic amines with a specific spatial relationship to a hydrophobic group, are associated with hERG liability across many chemotypes.
The standard guidance is to flag any compound with a predicted hERG IC50 below 30 micromolar. That guidance is appropriate for final compound selection. It is too conservative for early fragment-to-lead prioritization, because many structurally tractable intermediates pass through a hERG-flagged region of chemical space on the way to a clean final candidate, and excluding them entirely from SAR exploration forecloses synthetic routes that could otherwise be optimized.
For fragment expansion, we treat hERG flags as a structural design note rather than a hard exclusion criterion: which functional group is likely carrying the hERG liability, and is there a modification that would reduce the predicted affinity for the hERG channel while retaining the binding interaction you care about? A flagged compound with an interpretable structural basis for the hERG prediction is a better SAR candidate than a clean compound with an unexplained activity profile.
Weighting ADMET properties in Alkira's tox score
The composite tox score Alkira reports during fragment expansion is a joint signal across the full ADMET property set. It does not know your target class. The responsibility for interpreting that signal through the lens of your specific program biology sits with you.
What the score gives you is a consistent, property-specific breakdown that you can filter and weight manually. If you are running a CNS program, you can sort by the Pgp efflux prediction and treat any compound with a predicted efflux ratio above 2 as a hard flag regardless of its composite tox score. If you are running a gut-restricted target, you can ignore the permeability flag entirely and focus on the metabolic stability prediction in intestinal microsomes.
The composite tox score is a starting point for the conversation, not the end of it. The end of it is a per-property evaluation calibrated to the target biology you are actually working on.