AI Meets Expertise: A Hybrid Approach to Modern Target Identification

This article is adapted from a webinar featuring Daniel Bakowski, PhD (Sygnature Discovery), Saurav Bhaskar Saha, PhD (Sygnature Discovery), Iman Bhattacharya (QIAGEN Digital Insights), and Olivia Alder (QIAGEN Digital Insights). Learn more about the experts

Prefer to watch rather than read?

Access our on-demand webinar here.

The Real Challenge Isn’t Generating Targets, It’s Choosing the Right Ones

Target identification has always been complex. With the rise of AI, it has become significantly faster, but not necessarily easier.

AI can now generate hundreds, even thousands, of potential targets. On the surface, this feels like progress. But in reality, it introduces a new challenge: how do you confidently select the few targets worth investing time, budget, and experimental effort into?

As discussed in this conversation, the bottleneck in modern drug discovery is no longer idea generation. It is decision-making.

Why AI Alone Isn’t Enough

AI-driven platforms, particularly those leveraging causal analysis and multi-omics integration, are powerful tools for identifying potential targets. But they are not infallible.

In the webinar, Sygnature Discovery’s Daniel Bakowski shared a practical example from a CNS target discovery project. An AI-generated target appeared highly compelling: it was druggable, supported by strong network analysis, and aligned with the intended biology.

However, deeper expert review revealed a critical issue: while the target was relevant in the disease tissue, it was also highly active elsewhere in the body, introducing significant safety risks.

The AI wasn’t “wrong”, but it lacked context.

This highlights a key limitation: AI struggles to fully capture biological nuance, particularly when it comes to tissue specificity, systemic effects, and clinical feasibility.

From Target Generation to Target Deprioritisation

One of the most interesting shifts discussed was the move from target prioritisation to target deprioritisation.

Rather than asking “Which targets should we pursue?”, teams increasingly ask: “Which targets can we confidently rule out?”

Sygnature Discovery’s Saurav Saha described this as building a “target universe”, casting a wide net, then systematically filtering candidates using layered analysis, expert input, and iterative validation.

This approach reflects a broader shift in mindset:

Generate broadly
Evaluate rigorously
Eliminate early

Or, as Saurav put it: “Fail early. Exit early.”

The Power of a Hybrid Approach

What emerges from this discussion is not an AI-driven workflow, but a hybrid model.

In this model:

AI excels at scale: ingesting data, identifying patterns, and integrating multiple data types
Human expertise provides context: interrogating biology, assessing feasibility, and challenging assumptions

Crucially, this is not a linear handoff. It is an iterative, multi-disciplinary process involving:

Bioinformaticians
Biologists
Chemists
DMPK specialists
Translational scientists

Each discipline contributes a different lens, and importantly, each has the ability to veto a target.

Explainability: More Than Just Transparency

“Explainability” is often discussed in the context of AI, but in drug discovery it takes on a very practical meaning.

For scientists at the bench, explainability is not just about understanding how a target was generated. It is about biological plausibility.

Key questions include:

Does the biology make sense?
Are we interrogating the right pathway?
Do experimental systems reflect meaningful disease mechanisms?

As Daniel highlighted, confidence comes from building a coherent chain of evidence, not from a single data point.

And importantly, that chain can also be broken, leading to early and informed deprioritization.

Resolving Conflicting Data: The Role of Pathway Analysis

Conflicting datasets, particularly across different omics layers, are a common challenge.

For example, RNA and protein data may appear to point in opposite directions. Rather than treating this as a contradiction, QIAGEN’s Olivia Alder highlighted the importance of looking beyond individual data points to pathway-level effects.

Tools like Ingenuity Pathway Analysis (IPA) enable researchers to:

Integrate multiple data types
Map signals onto biological pathways
Identify convergence at a functional level

This more holistic view often reveals that what appears to be conflict at the gene level may actually align at the system level.

However, even with advanced tools, human expertise remains essential:

To validate data quality
To interpret results in context
To align findings with the end goal of the program

From Targets to Pathway

Another key insight is that drug discovery often moves beyond individual targets.

Initial analysis may highlight a specific protein, but practical considerations—such as tractability or safety—can shift focus toward the broader pathway.

This creates new opportunities:

Alternative nodes within the pathway
Different mechanisms of action
Safer or more tractable intervention points

In some cases, the most promising strategy is not targeting the original protein, but modulating the pathway in a different way.

Where AI Adds the Most Value Today

While AI has limitations, its strengths are clear.

Across the discussion, one area stood out: multimodal data integration.

AI enables teams to:

Combine diverse datasets
Identify patterns across domains
Generate structured hypotheses more efficiently

As Saurav described, it acts as a powerful assistant, helping teams move faster from raw data to a defined problem statement.

Looking ahead, there is also growing interest in using AI to extract subtle signals from complex datasets, such as clinical or longitudinal data, to uncover new opportunities for drug repositioning or novel target discovery.

What Are We Still Missing?

Despite these advances, an important question remains:

What are we overlooking?

AI can generate unexpected and novel targets, but these are often deprioritised due to:

Lack of existing evidence
Higher perceived risk
Limited tractability

As Daniel noted, there is always a tension between pursuing well-supported targets and exploring more speculative biology.

This raises a broader challenge for the industry:
how to balance risk, novelty, and feasibility without missing future breakthroughs.

Practical Takeaways for Target ID Teams

The discussion closed with a set of practical recommendations:

Design workflows to detect failure early
Don’t optimise for success signals, identify reasons to stop
Invest in data quality
Clean, well-curated data underpins every decision
Look for unexpected signals
Apparent risks or anomalies can represent hidden opportunities
Stay focused on context
Biology, modality, and program goals must guide interpretation

Final Thought

The role of AI in drug discovery is no longer in question.

The real challenge is how to use it effectively in practice.

As this conversation makes clear, the most successful approach is not AI replacing human expertise, but AI and experts working together, combining scale with context, and speed with judgment.

In an increasingly complex discovery landscape, this hybrid model is not just advantageous, it is becoming essential for making faster, more confident decisions about which targets truly matter.

Learn more about Sygnature’s AI-enabled drug discovery capabilities here.

The Panelists of the Webinar