Antibody Screening – Out of billions of molecules, which ones are worth keeping?

The full chain of antibody discovery can be roughly divided into two stages: the first is 'generation'—producing 10⁹ to 10¹¹ antibody variants through immunization, library construction, or computational design; the second is 'screening'—gradually weeding out candidates from this diversity until only a few remain.

These two stages don’t hold the same strategic weight. The core investment in the generation stage is platform setup and library construction, which are essentially fixed costs—you can reuse the system once it’s built. Screening, on the other hand, is a variable cost for each project—each additional round consumes more time, materials, and manpower. More importantly, every decision in the screening phase irreversibly eliminates candidate molecules: clones that are discarded never return to the system, and epitopes that are missed are hard to rediscover later.

The word 'screening' itself is misleading; it suggests a passive, neutral process—like a sieve, where particles smaller than the holes drop through and bigger ones stay. But antibody screening is never entirely neutral: each round’s conditions—antigen format, washing strength, detection method—actively define 'what a good antibody is.' This definition includes both the evaluation criteria set by researchers and systematic biases imposed by current technology, jointly shaping the boundary of final candidates. Screening doesn’t find the best molecules in the natural sequence space; it finds the best molecules within the pre-defined standard system.

This is the first principle of antibody screening: the screening conditions are the definition.

1. Antigen Format: The First Source of Bias in the Screening Funnel

The first key decision in antibody screening is "in what form to present the target antigen." This decision is usually made in the first few weeks of the project launch, yet it influences the direction of all subsequent screening rounds.

Recombinant proteins are the most commonly used antigen strategy: the purified extracellular domain protein can be directly coated on an ELISA plate or biotinylated and fixed on streptavidin magnetic beads. This approach is clean, low-background, and high-throughput. But the key assumption of this strategy — "recombinant extracellular domain proteins can equivalently represent the cell-surface native protein" — doesn’t hold up in many scenarios. A classic failure mode is that antibodies identified show strong signals against recombinant protein in ELISA but fail to effectively bind the native target on cells, with weak FACS verification signals. The underlying reasons fall into two categories: first, proteins may partially denature or orient randomly when passively adsorbed to polystyrene surfaces, causing some conformational epitopes to be lost or new ones exposed; second, purification can strip away transmembrane anchors and natural lipid microenvironment support, altering conformational features.

This bias is not an intentional choice by researchers but a trade-off under current technical constraints: high-throughput screening systems can hardly maintain the complete native conformation of membrane proteins at low cost, so recombinant proteins are a compromise between throughput and cost.

Whole-cell screening preserves the native conformation and post-translational modifications of antigens, theoretically closer to the in vivo situation, but comes at a significant cost. A full cell surface has thousands of proteins, each a potential nonspecific binding site, creating high background noise that makes it harder to distinguish low-abundance targets or moderate-affinity specific binders. A subtler problem is that the density and distribution of target antigens on commonly used engineered cell lines systematically differ from expression in human tissues under pathological conditions, which may cause antibodies identified to reveal issues in later tissue cross-reactivity tests.

Screening for multi-pass membrane proteins (such as GPCRs and ion channels) presents unique challenges: the conformation of extracellular loops is determined by the overall arrangement of transmembrane helices, which is easily lost during extraction and purification. Moreover, these targets exhibit significant conformational differences in different activation states; if the screening antigen is fixed in a single conformation, the resulting antibodies may not bind the physiologically relevant conformation in vivo. Currently, industries use strategies like nanodiscs, virus-like particles (VLPs), and liposomes to mimic the natural membrane environment. Each method balances "conformational fidelity" with "screening throughput": the higher the conformational fidelity, the more complex the system and the lower the achievable throughput; the higher the throughput, the more simplified the antigen format, making it harder to guarantee conformational fidelity.

Choosing the antigen format essentially sets a bias for the entire screening process — the screening system naturally favors antibodies that recognize this specific form of antigen. All subsequent screening rounds reinforce this pre-set bias, rather than actively correcting it.

2. Screening Throughput: Characterization Capability Determines Funnel Resolution

Selection funnel

The first stage of the screening funnel is physical slotting or FACS sorting: after 3~4 rounds of bacteriophage selection, the total clonal diversity is compressed from the initial 10⁹~10¹¹ to the 10⁴~10⁵ level; Yeast or mammalian cell display combined with multiple rounds of FACS sorting can gradually compress diversity to the scale of hundreds of cells. This stage deals with group-level enrichment behavior, and the throughput bottleneck is not prominent.

The real bottleneck appears in the second stage: from the enriched clone database, select 10²~10³ monoclones for characterization one by one.

In this step, each candidate clone must complete a set of basic validations: expression supernatant preparation, ELISA binding validation, FACS cell-level binding confirmation, and preliminary epitope boxing. Experiments such as ELISA and FACS can be completed in batches using 96-well / 384-well plates, offering relatively high operating throughput; But the real rate limiting step comes from deeper characterization: for the top 20~30 clones, protein purification, BLI/SPR binding kinetics measurement, SEC-HPLC aggregate ratio detection, and so on. These experiments rely on precision instruments and purification processes. Monoclonal detection cycles are long and labor costs are high, making them the core bottleneck for flux characterization.

The upper limit of characterization capability determines the final resolution of the screening funnel. If the platform can consistently characterize more than 200 candidates, the filtering strategy can adopt coarser granularity—aggressive elimination and rapid convergence; If characterization capability can only cover 50 clones, screening must adopt a more conservative strategy—retaining more candidates each round to prevent accidental kills, which also extends project timelines.

Under ideal conditions of clear targets and good antigen quality, antibody discovery projects in the industry typically require 6~12 months from immunization/selection to candidate molecular targeting; For projects with difficult targets or requiring deep engineering, the period may be extended to 18~24 months, covering the entire process including target validation, antigen preparation, library selection, molecular characterization, and lead function validation. The core reason for the slow progress is not the low efficiency of diversity generation, but the limited throughput of monoclonal depth characterization, which is one of the most critical speed-limiting steps in the entire process.

3. Tabletop boxing: the core decision dimension that is postponed

In the usual screening process, epitope information is typically analyzed systematically only in the later stages of a project: first, a few candidates with the highest affinity are selected, and then their binding epitopes are determined through competition experiments or structural analysis. This serial process of 'screen for affinity first, determine epitope later' carries an implicit assumption: that affinity and epitope are statistically independent, so selecting based on affinity first won't systematically exclude antibodies targeting specific epitopes.

However, in engineering practice, this assumption doesn’t hold for all targets. If a functional epitope (like a receptor's ligand-binding site) happens to be located in a recessed region of the protein structure, the area available for CDR contact is limited — with typical antibody paratope requirements (about 700–900 Ų), a narrow pocket might restrict both the contact density of complementary residues and the freedom for mutations, meaning that the theoretical affinity limit could be lower than for epitopes on flatter surfaces. This situation is more common for special targets like GPCR extracellular loops or enzyme active sites. If the first round of screening uses a 'top 10% affinity' hard cutoff, antibodies that bind the functional epitope but have slightly lower affinity might be eliminated early — not because they’re 'undruggable', but because they weren’t evaluated by the right standard.

It’s worth noting that this rule isn’t universal. For most targets, like viral spike proteins or cytokine receptors, strong functional neutralizing antibodies usually bind the ligand-binding sites or other functional epitopes with very high affinity. Whether a 'functional epitope shows lower affinity' situation occurs ultimately depends on the structural characteristics of the target protein.

Epitope binning — grouping candidate antibodies based on binding epitopes through competition experiments — is a necessary method to address this issue. But in traditional workflows, binning is usually done after the candidates are locked down, serving only as supplementary information, by which point it’s already lost its decision-making value in the screening stage. Moving epitope binning to the early stages of screening, even at a coarser level (like simply distinguishing ‘ligand-competitive’ vs. ‘non-ligand-competitive’), can effectively preserve epitope diversity and prevent the end of the screening funnel from ending up with 'all candidates hitting the same immunodominant epitope'.

Concave functional epitope vs Flat immunodominant epitope

4. Developability: Drug-Likeness Risks Delayed Until Screening Stage

The developability risks of antibodies—such as aggregation tendency, low pH stability, deamidation risk, oxidation sensitivity, and insufficient solubility—are major reasons for delays and failures in the CMC stage. These risks are embedded in the antibody sequences and can actually be assessed during the discovery phase, but they are often not included in the priority ranking system during early screening.

The reason behind this is not hard to understand: a full developability assessment traditionally involves low-throughput experiments that require purifying milligram amounts of antibody protein, then analyzing them one by one using instruments like analytical HPLC, DSC, or DSF. When the number of candidate clones is still in the hundreds, performing a complete developability assessment for each candidate is impractical in terms of time and cost.

However, this doesn’t mean developability can’t be considered during the screening stage. At the sequence level, you can directly identify post-translational modification hotspots (like Asn-Gly deamidation sites, Met oxidation sites), unpaired cysteines, surface charge distribution, isoelectric points, and other risk factors. Combined with structural prediction, you can further evaluate the surface exposure of hydrophobic residues in the CDR regions. These assessments don’t require protein expression or purification and can provide an initial prediction straight from the sequence, serving as a reference dimension for prioritizing clones.

Here’s a specific scenario: a clone ranked 15th in affinity but with very low sequence-structure risk factors; and another clone ranked 3rd in affinity but with a high proportion of surface-exposed hydrophobic residues in the CDR-H3 region—in a traditional "affinity-only ranking" logic, the latter would be prioritized. But with developability pre-screening included, the priority of the former would be reassessed.

It’s important to note that sequence-structure-based developability predictions have some false positives and negatives, and can only serve as an auxiliary reference for prioritization, not as a strict standard for eliminating clones directly. Antibody aggregation is influenced by multiple factors; the exposure of hydrophobic residues in the CDR is highly correlated with aggregation tendency and is an important factor, but not the only determinant. Experimental validation is still ultimately required.

5. MatwingsVenus™ Intelligent Agent: Turning Screening from "Passive Elimination" to "Active Optimization"

The underlying logic of traditional antibody screening is "rank by a single affinity metric and cut off the rest." The core value of the MatwingsVenus™ (Xiaowu™) intelligent agent is that it introduces multi-dimensional information into the screening stage, upgrading the decision-making on "who to eliminate and who to keep" from single-metric ranking to active optimization based on multi-objective trade-offs.

Conventional single-parameter ranking vs Multi-parametric integrated ranking

Conformational Validation During Antigen Design

Before expressing the antigen, the MatwingsVenus™ (Xiaowu™) agent can analyze the 3D conformation of the target protein through structural prediction: whether the boundaries of transmembrane regions are clear, whether truncation designs might mistakenly expose hydrophobic surfaces, and the location and spatial shielding of key glycosylation sites. This information can help researchers evaluate the conformational fidelity of "recombinant proteins as screening antigens," optimize antigen truncation and modification designs, and reduce the risk of conformational bias in the first round of screening right from the start. It’s important to note that structural prediction can help lower bias but cannot completely eliminate inherent differences between recombinant and natural proteins.

Dynamic Adjustment of Screening Pressure

Based on the structural features of the antigen surface, the protein function prediction module can preliminarily assess the theoretical accessibility of different epitope regions: which regions are more likely to be contacted by antibody CDRs and which regions may not stably present due to conformational flexibility. This information doesn’t directly predict which epitopes will generate high-affinity antibodies but can help researchers decide whether to adjust the screening strategy across multiple rounds (e.g., adding ligand competition elution, using conformationally locked antigens) to protect candidates binding specific functional epitopes and avoid systematic dilution by high-affinity immunodominant epitopes.

Sequence-Structure Integrated Pre-screening for Developability

Before candidate sequences are finalized, the MatwingsVenus™ (Xiaowu™) agent can use sequence analysis and structural prediction to label risk factors such as hydrophobic patch exposure in CDRs, post-translational modification hotspots, and surface charge distribution. For sequences with risks, it can also recommend conservative mutation schemes to reduce sequence risk without affecting antigen-binding potential. This upfront evaluation shifts the traditional "screen for affinity first, evaluate developability later" sequential process into a "simultaneous evaluation and combined ranking" parallel workflow.

Meanwhile, priority ranking based on multi-dimensional scoring allows limited deep characterization resources to be focused on high-potential clones first, improving the effective use of characterization resources and indirectly alleviating bottlenecks caused by limited throughput.

Conclusion

On the surface, antibody screening seems like a process of elimination—remove the bad and keep the good. But what defines "good" is preset from the very start by antigen format, detection methods, and ranking logic.

The truly excellent screening strategy is not about more strictly following preset standards but constantly reviewing those standards: Are the molecules we screened really the ones we need? When deviations in antigen format can be corrected by structural prediction, when epitope diversity can be protected by binning strategies, and when developability risks can be preemptively flagged at the sequence level—the screening process is no longer a passive elimination funnel but an active optimization system.

The ultimate goal of screening has never been to find "the molecule with the highest affinity" but to identify "the molecule most worth advancing" while there’s still time to make a choice.