Prioritizing Bispecific ADC Targets: From Complex Biology to Evidence-Based Decisions

Having an abundance of targets does not seem like a problem but for resource constrained biotech start-ups, deciding where to commit resources for experimentation can be a major bottleneck to progress towards the clinic. In the emerging space of antibody drug conjugates, a team may have several validated and emerging targets and an established ADC development approach, yet the number of possible target pairs can quickly become difficult to evaluate systematically. Each pair brings its own considerations around tumor biology, co-expression, safety, molecular developability, and clinical relevance.

For one ThinkBio.Ai client, this was exactly the challenge. Their NSCLC program had an established ADC platform and a broad portfolio of potential targets but needed to determine which bispecific combinations warranted deeper investigation. ThinkBio.Ai® used DrugSuccess.AI® to evaluate the target landscape across multiple evidence dimensions and narrow a broad combinatorial space into a focused set of candidates for further consideration.

Why is target selection important for bispecific ADC development?

Target selection is particularly important in bispecific ADC development because the therapeutic concept depends on the relationship between two targets rather than the properties of a single antigen.

Their expression across tumor and normal tissue, co-expression within relevant cell populations, biological relevance, and potential contribution to internalization and payload delivery all influence the rationale for pursuing a particular combination.

As the number of potential targets grows, the challenge becomes more than identifying individual antigens with attractive characteristics. Development teams need to understand which targets make sense together and which combinations have enough supporting evidence to justify further work.

The broader ADC field has been expanding rapidly. A 2024 review reported that 10 bispecific ADCs were undergoing clinical trials at the time, while highlighting target selection, antibody format, and linker-payload design as important considerations in their development.

How do you start prioritizing bispecific ADC target pairs?

Bispecific ADC target pairs can be prioritized by evaluating the individual targets first and then examining their combined biological and development characteristics. Relevant evidence can include tumor and normal tissue expression, cellular co-expression, biological relevance, safety considerations, molecular developability, indication context, and potential biomarker strategies.

For the NSCLC program, the analysis began with a broad set of candidate targets. ThinkBio’s team did the initial assessment of expression patterns across tumor and normal biological contexts, including high-resolution cellular analyses.

The objective was to understand which targets had sufficient tumor association and biological relevance to warrant consideration as part of a bispecific design.

This first stage was important because pairwise evaluation becomes considerably more useful once the individual target landscape has been narrowed. Rather than treating every theoretical combination equally, the analysis created a more focused set of targets for deeper comparison.

The biological context of the targets also matters. For a bispecific ADC, co-expression of both targets within the same cellular environment can support the rationale for dual targeting, particularly where the intended mechanism depends on binding and internalization of the resulting complex. A target that looks promising in isolation may therefore warrant a different assessment when considered alongside a second antigen.

How did we use DrugSuccess.AI to prioritize bispecific ADC targets?

AI supports bispecific ADC target prioritization by bringing multiple biological and molecular evidence sources into a common analytical process. In the NSCLC case, ThinkBio’s DrugSuccess.AI solution was used to examine target characteristics across different biological contexts before evaluating pairwise combinations. This allowed the analysis to move beyond a single expression measurement and consider how the evidence converged around each potential candidate.

A useful target prioritization process needs to consider more than tumor expression. For bispecific ADCs, relevant factors include tumor and normal tissue distribution, co-expression within relevant cellular populations, biological rationale, predicted safety, molecular developability, indication relevance, and the potential to support a practical biomarker strategy.

For the NSCLC program, shortlisted targets were evaluated as pairs using an integrated assessment of efficacy potential and safety considerations. First expression filter identified 8 targets which can generate 28 different unique heteromeric combinations and by using other filters DrugSuccess.Ai was able to reduce it to 1 bispecific – a 28 times enrichment. Among the combinations assessed, CEACAM5–MET emerged as a leading candidate based on tumor relevance and predicted safety. Further analysis showed that the two targets were co-expressed within tumor-associated epithelial populations at substantially higher levels than in normal tissue.

This provided several converging lines of evidence for the pair. The targets were relevant to the tumor context, their co-expression supported the proposed bispecific design, and the predicted safety profile provided another dimension for prioritization.

From here scientists can determine where experimental attention may be most useful. That distinction becomes increasingly relevant as therapeutic design becomes more complex. A computational assessment can organize evidence and identify patterns, while laboratory and clinical studies remain necessary to establish whether a candidate performs as expected.

How did ThinkBio.AI extend analysis beyond tumor biology?

A promising target pair still needs to make sense from a development perspective. Molecular properties, indication biology, patient selection, and biomarker feasibility can influence whether a candidate has a practical path toward further development.

DrugSuccess.AI® therefore assessed translational feasibility alongside target biology. Molecular developability was examined using predictive models covering physicochemical, structural, and immunological attributes, with molecular and sequence-derived features used to identify potential risks and optimization considerations.

The analysis also examined target expression across clinically relevant molecular subgroups of NSCLC. The prioritized targets showed consistent expression across subsets associated with common oncogenic drivers, providing additional context for considering the candidate in patients whose disease may have progressed following existing targeted therapies.

Spatially informed analysis provided another layer of context by examining where target-expressing cells occurred within the tumor microenvironment. In the case study, these cells were observed in regions associated with immunosuppressive characteristics. Such observations can help researchers consider how target biology relates to different tumor states and potential patient populations.

This broader assessment helps connect target selection with the development questions that follow it. A target pair needs a biological rationale, but it also needs to fit within a plausible molecular and clinical development strategy.

What does this mean for bispecific ADC development?

Bispecific ADC development brings several interconnected decisions into a single development pathway. Target selection needs to account for the biology of two antigens, their expression within relevant cellular environments, potential safety considerations, and the molecular and clinical characteristics that could influence downstream development.

The NSCLC case demonstrates how DrugSuccess.AI® can support this process by connecting these evidence layers during early target prioritization. The analysis moved from a broad portfolio of potential targets to a focused set of combinations and identified CEACAM5–MET as a prioritized candidate based on converging biological and development considerations.

For teams evaluating increasingly complex therapeutic designs, this creates a more structured starting point for experimental work. The objective is not to make the development decision through AI alone, but to give scientists a clearer evidence base from which to determine which candidates merit deeper investigation.