This is the final post in a series reflecting on themes from FOG (Boston) and EACR (Budapest).

Click to read Part One, Part Two and Part Three.

I’ve saved this topic for last because it’s less of an observation and more of a personal perspective, one I’d happily discuss over a coffee.

Across both conferences, the direction was clear: researchers want more data types from more compartments, using less of their precious samples. Multiomics integration is the expectation. The technology to deliver on that expectation is maturing rapidly.

But sitting through multiple sessions on multiomic workflows, I kept thinking about a distinction that didn’t always get enough airtime: describing a system in ever-greater molecular detail is not the same as understanding it.

The papers that ultimately change clinical practice tend not to be the ones that profiled fifteen layers and presented a correlation heatmap. They’re the ones that identified a specific mechanism, validated it rigorously, and translated that understanding into something actionable, whether that’s a diagnostic marker, a therapeutic target, or a clinical decision.

Multiomic profiling is enormously valuable as a discovery tool. It’s how you find the signal. But the signal still needs to be followed up with focused, mechanistic work that tests whether the correlation is real, whether it’s causal, and whether it holds up across patient populations and experimental conditions.

I think the field sometimes risks conflating the ability to measure more with the ability to understand more. The distinction matters most when the goal is clinical translation.

This tension showed up concretely in the spatial proteomics landscape. As I discussed in Part 2, the field is bifurcating into high-plex discovery panels and low-plex clinical deployment panels. That bifurcation is essentially the field recognising this point: discovery needs breadth, but clinical translation needs depth.

Once you’ve used a hundred-marker panel to identify the seven targets that matter in a given tumour microenvironment, what pharma actually wants is a small, validated, reproducible panel that reads out those seven targets reliably across multiple clinical samples and sites. The value shifts from “how many markers can we measure” to “how well do we understand the ones that matter.”

That transition from discovery to deployment is where mechanistic understanding becomes essential. You can’t validate a clinical panel without understanding why those markers matter, how they interact, and what confounders might affect interpretation in a real-world clinical setting.

In this series we covered multiomic convergence, spatial maturation, AI integration, the shift to human-relevant models, and now the case for mechanistic depth over breadth. These themes aren’t independent. They’re all aspects of a field that’s becoming more integrated, more translational, and more demanding in terms of rigour and reproducibility. 

The decisions being made now about standards, validation, and infrastructure will determine how quickly these technologies reach routine clinical use. As I mentioned in Part 1, the scientists who shaped this field thought in decades. The challenges ahead deserve the same approach. 

At Millennium Science, we attend conferences like FOG and EACR to understand where the field is heading, so we can make sure ANZ researchers have access to the tools and technologies that support their best work. If any of these themes resonate with your own research, we’d genuinely welcome the conversation.

This is the third post in a series reflecting on themes from FOG (Boston) and EACR (Budapest). This post covers two themes that were less flashy than multiomics and spatial biology but are potentially more consequential in the long run: the integration of AI across the field, and a structural shift in the experimental models researchers are using. Both are less about new science and more about the shifting infrastructure underneath. 

Click to read Part One and Part Two.

At FOG in particular, AI wasn’t confined to its own dedicated track. It was woven through virtually every session, across genomics, proteomics, drug discovery, pathology, and clinical data science. Assay automation, foundation models, agentic platforms, generative tools, and machine learning pipelines all featured across the full breadth of the programme. 

The signal isn’t that AI is coming to biology. The signal now is that AI is assumed. It’s becoming infrastructure rather than innovation. The more interesting questions are about what AI currently can’t do well, and where the field’s enthusiasm is running ahead of its validation. 

The hardest of those questions, and the one I kept hearing across both conferences: QC and standardisation of multi-modal data remains a genuine bottleneck. Sequencing data has well-established quality metrics (Q scores, coverage depth, error rates) that the field has spent two decades refining. High-content spatial imaging data doesn’t have equivalent consensus standards. Batch effects, optical artefacts, tissue quality variation, and registration errors all introduce noise that current QC frameworks handle inconsistently. 

This matters because AI tools are only as good as the data they’re trained and validated on. If the underlying data quality standards aren’t solved, AI amplifies the noise rather than cutting through it. Whoever solves the reproducibility and validation problem at scale for spatial and multimodal imaging data will capture disproportionate value as the field translates into clinical use. 

Less visible than AI and multiomics, but very recognisable especially at EACR is that research is moving away from animal models toward more human-relevant systems. Organoids, in vitro tissue models, and banked human tissue are gaining ground as primary experimental targets. 

This shift is being driven from multiple directions. Regulatory frameworks are increasingly permitting non-animal alternatives for preclinical testing, and the economics favour it. Animal studies are slow, expensive, and have well-documented translational failure rates. And the scientific argument is straightforward: if the goal is to understand human biology, starting with human material makes more sense than starting with a model organism and hoping the biology translates. 

The technology ecosystem around this shift is responding faster than I’d expected. Plenty of new platforms for 3D cell culture, tissue engineering, and high-throughput organoid screening were visible across both conferences, from established companies and newer entrants alike. The infrastructure to support human-relevant preclinical research is being built now, and the pace suggests this is a structural shift rather than a passing trend. 

These two themes (AI integration and the move to human-relevant models) might seem unrelated, but they share a common thread: both are about the foundations of how research gets done changing underneath the science itself. AI is reshaping the data infrastructure. Organoids and banked tissue are reshaping the experimental infrastructure. Together, they’re redefining what a well-equipped lab looks like.

Interested in how these infrastructure shifts connect to your lab’s workflows? Whether it’s spatial data analysis, 3D cell culture, or automation, we’re here to help you navigate what’s next.

This is the second post in a series reflecting on themes from the Festival of Genomics (Boston) and EACR (Budapest).

In this post, I want to unpack the two themes that dominated the scientific sessions at both conferences: the convergence of multiomic workflows, and the maturing spatial biology landscape.

The most consistent theme across both conferences. Researchers are looking for multiple molecular readouts from the same sample, and the general approach is shifting to expect it.

From the genomics side, sequence is increasingly a starting point rather than a destination. Methylation status, three-dimensional chromatin architecture, and RNA modifications are now treated as essential layers that sit alongside sequences. The range of assay chemistries addressing these layers has expanded noticeably, and the expectation is that a complete genomic picture includes epigenetic context, not just the linear sequence of bases.

At the sequence level itself, whole-genome sequencing (ideally diploid and phased) is slowly becoming the expectation over exome as the cost differential narrows. The information lost by sequencing only coding regions is becoming harder to justify when intronic, regulatory, and structural variant information is increasingly recognised as clinically relevant.

In addition to genomics and transcriptomics, the shift in focus to include proteins is becoming increasingly important, as proteins are the actual functional units of biology. As such, alternative top-down proteomics approaches are complementing traditional mass spectrometry, opening protein profiling to labs without dedicated proteomics infrastructure. Beyond proteins themselves, the field is increasingly profiling metabolomics and lipidomics alongside traditional proteomics, building toward a more complete molecular phenotype.

The boundaries between different biological compartments are blurring. Workflows that were previously siloed are converging, driven by both the scientific recognition that biology doesn’t operate with disciplinary boundaries, and the practical reality that sample material is often limited. The goal is to extract as many meaningful readouts as possible from each sample.

Spatial transcriptomics and proteomics have been conference staples for several years, but the conversation at both FOG and EACR had clearly matured past raw capability and into practical deployment. 

The spatial proteomics field in particular appears to be bifurcating. High-plex panels (10s to 100s of markers) are doing discovery work, casting a wide net to identify which proteins matter in each tissue context. But lower-plex panels (<10 markers) are emerging as the format that pharma and clinical labs actually want for deployment. A small, validated, reproducible panel gives a clearer answer in a clinical workflow than a hundred-marker discovery panel does.  

Spatial methods more broadly are moving from research toward translational and clinical use, but the view from pharma is more measured than the hype. Several pharma-side speakers noted that spatial technologies validate digital pathology rather than replacing it, H&E staining remains the clinical gold standard, and that the current platforms aren’t quite mature enough for routine diagnostic use. 

The structural gaps everyone points to are remarkably consistent: automation (too much hands-on time per sample), standardisation (too much variability between operators and sites), and analysis tools that hold up under industry reproducibility requirements. 

None of this means spatial is overhyped. It means the field is in the transition phase between research tool and clinical infrastructure, which is the most interesting and consequential phase to be in.

We’d be happy to discuss how these trends connect to the tools and platforms available in ANZ.

Hi, I’m Gerry Ma, Technology and Development Manager here at Millennium Science. I joined Millennium Science a few months ago, having come from commercial roles across the flow cytometry, single-cell genomics, and spatial transcriptomics fields. I now spend a good portion of my time talking to scientists and technology developers about where the field is heading and what tools ANZ researchers need to do their best work. 

In early June, I attended two conferences back-to-back: the Festival of Genomics (FOG) in Boston, and the European Association for Cancer Research congress (EACR) in Budapest. Two very different audiences (genomics and AI in Boston; translational cancer research in Budapest), but the overlap in themes was more striking than the differences.

This is the first in a short series of Field Signals blogs where I’ll share what stayed with me. Not a comprehensive conference review, but the observations I kept coming back to upon returning home, and the ones I think matter most for how research is going to look over the next few years. 

Before diving into the technical themes, I wanted to share one personal highlight that set the tone for everything else. 

At FOG, George Church (Harvard Medical School), Mark Adams (The Jackson Laboratory for Genomics Medicine), and Sorin Istrail (Brown University) discussed the future of genomics in a session that also served as a tribute to the late J. Craig Venter, who was originally meant to speak. Their conversation ranged across multiomics, in situ genomics, xenotransplantation, national-scale sequencing programmes, and the role of AI in biological data at population scale. 

But what stayed with me was simpler than any of those topics. Watching three scientists who shaped the field reflect on the history of genomics and past conversations with Craig Venter was a reminder that the technologies we now take for granted were built by people who had genuine scientific progress as their primary motivation, and who thought in decades rather than quarters. 

That framing stuck with me through both conferences. The decisions being made right now about how we integrate multiple data types from different platforms, how we standardise methods, and how we shift toward new experimental models are the decisions that will define the trajectory of the next decade of biological research. These decisions deserve the same kind of long-term thinking.

Over the next few posts, I’ll unpack the themes that came through most clearly: 

• Part 2 will cover the convergence of multiomic workflows and the maturing spatial biology landscape. 

• Part 3 looks at AI integration and a quieter but potentially more consequential shift in experimental models. 

• Part 4 wraps up with a personal perspective on why mechanistic depth still matters more than breadth.