Developability assessment asks one question before you commit to a lead: will this molecule survive expression, purification, formulation at high concentration, storage and administration without a property forcing you back to the bench? Eight dimensions cover most of what goes wrong, and all eight can be screened from the sequence and a structural model before any material exists. The output is not a verdict. It is a ranked list of candidates with the specific risk each one carries and the specific experiment that would confirm or clear it.
The eight dimensions
| # | Dimension | What goes wrong | Typical sequence or structure signal |
|---|---|---|---|
| 1 | Aggregation and self-association | Particles, low recovery, immunogenicity risk | Large hydrophobic patches, exposed aromatic clusters in the CDRs |
| 2 | Viscosity at high concentration | Subcutaneous formulation above about 100 mg/mL becomes unmanageable | High net charge asymmetry between VH and VL, charged patches |
| 3 | Solubility and opalescence | Phase separation, filtration problems | Low net charge near the formulation pH, high surface hydrophobicity |
| 4 | Charge heterogeneity and pI | Purification and ion exchange behaviour, isoelectric precipitation | Theoretical pI close to the intended formulation pH |
| 5 | Chemical liabilities | Deamidation, isomerisation, oxidation, clipping during storage | NG, NS, DG, DS motifs in CDRs, exposed methionine and tryptophan |
| 6 | Immunogenicity risk | Anti-drug antibodies, loss of exposure | Non-germline residues, predicted MHC-II binding peptides |
| 7 | Expression and manufacturability | Low titre, unpaired cysteines, glycosylation in the wrong place | Free cysteines, N-glycosylation sequons in CDRs, VH-VL pairing mismatch |
| 8 | Conformational and colloidal stability | Low melting temperature, pH sensitivity during low-pH viral inactivation | Unusual framework, destabilising CDR grafts |
What each dimension costs if you miss it
The reason to screen early is not elegance, it is the cost of discovering the problem late:
- A viscosity problem found in formulation development forces a change of route (intravenous instead of subcutaneous), a device change or a reversion to an earlier candidate.
- A deamidation hotspot in a CDR found during stability studies means either a sequence change, which restarts characterisation, or a cold chain and a short shelf life.
- A free cysteine found in process development shows up as a disulfide scrambling problem in every batch.
- An N-glycosylation sequon in a CDR becomes a lot-to-lot heterogeneity you have to control for the life of the product.
None of these is fatal on its own. All of them are much cheaper to address while you still have ten candidates than when you have one.
How to read the screening output
Four rules make the difference between a useful ranking and a false sense of security:
1. Rank, do not grade. The relevant information is which candidate is worse than which, not whether a score is "good". Absolute thresholds transfer poorly between antibody formats and between calculation methods. 2. Make the formulation explicit. Charge, solubility and viscosity risks depend on the pH and ionic strength you intend to use. The same molecule can be low risk at pH 5.5 and high risk at pH 6.5. 3. Separate liability from exposure. An NG motif in a framework loop that is buried behind the CDRs behaves differently from the same motif at the tip of CDR-H3. Structure, not just sequence, decides. 4. Carry the uncertainty. A prediction near a decision boundary is a reason to run the assay, not to decide. Say which candidates are confidently separated and which are not.
Which experiment clears which risk?
Each dimension has a cheap confirmatory assay, and the screening exists to decide which of them to run on which candidate.
| Dimension | First confirmatory experiment |
|---|---|
| Aggregation | SEC and dynamic light scattering after accelerated stress, polysorbate-free |
| Viscosity | Concentration series to the target, cone-plate or microfluidic viscometry |
| Solubility | PEG-induced precipitation or concentration to failure |
| pI and charge | Capillary isoelectric focusing, ion exchange profile |
| Chemical liabilities | Forced degradation plus peptide mapping, site-specific quantitation |
| Immunogenicity | Ex vivo T cell assay on a donor panel |
| Expression | Small-scale transient expression titre and product quality |
| Conformational stability | Differential scanning calorimetry or fluorimetry, plus a low-pH hold |
A practical order of work
For a panel of ten to thirty candidates, a workable sequence is: screen all candidates on all eight dimensions in silico, drop only the candidates with hard liabilities that cannot be engineered away in your timeline, then run the two or three cheapest assays on the remaining set, and use those results to calibrate how much weight the predictions deserve for this panel. By the time three candidates are left, every remaining question should be an experiment, not a model.
Limits of the statement
- Predictions from sequence and a homology model are triage, not characterisation. They prioritise work; they do not replace the assays above.
- Immunogenicity risk from predicted epitopes is the weakest of the eight. Clinical immunogenicity depends on dose, route, patient population and aggregation state, none of which is in the sequence.
- Developability behaviour is format-dependent. Rules learned on IgG1 do not transfer unchanged to bispecifics, fragments or antibody-drug conjugates, where linker and payload dominate.
- The formulation is a variable, not a constant. A candidate ranked low can become acceptable with an excipient change, which is a formulation study, not a screening result.
Frequently asked questions
What is developability assessment for antibodies?
A structured check, before lead selection, of whether a candidate can be expressed, purified, formulated, stored and administered without a molecular property blocking development. It covers aggregation, viscosity, solubility, charge, chemical liabilities, immunogenicity, expression and stability.
Can developability be assessed from the sequence alone?
Partly. Sequence reveals chemical liability motifs, free cysteines, glycosylation sequons, theoretical pI and germline distance. Aggregation, viscosity and solubility risks depend on surface properties and need at least a structural model, and all of it stays a prediction until the confirmatory assays run.
Which CDR motifs are chemical liabilities?
NG and NS for deamidation, DG and DS for isomerisation and clipping, exposed methionine and tryptophan for oxidation, and free cysteines for disulfide scrambling. Whether a motif matters depends on its solvent exposure and local flexibility.
At what concentration does viscosity become the deciding factor?
For subcutaneous administration the target is typically 100 mg/mL or above in a small volume, and that is where viscosity, self-association and opalescence separate candidates. Below roughly 50 mg/mL these risks rarely decide a lead.
How many candidates should enter a developability screen?
As many as you can sequence. The screen is cheap relative to any assay, so its value is highest when it is applied to the whole panel and used to decide where to spend experimental effort.