Tumour diversity is what makes cancer so difficult to treat, but scientists and doctors are starting to address the issue through the use of single cell technologies. Two emerging technologies, AI and multi-omics single cell measurements, will play a key role.
by Dr Andreas Schmidt
The words “precision medicine” and “personalized medicine” have appeared in headlines since the publication of the human genome, but while many advances have been made, relying solely on mutation detection has not led to the medical revolution that had been promised. We are now entering a new era of personalized medicine: Not only is each patient treated as having a unique disease, but we also must look at each tumour cell and its specific environment to guide treatment.
Personalizing Does Not Equal Precision
The reality, though, is still far from the ideal. Only a small percentage of patients are afforded some level of personalization beyond standard treatment. When they are, uncertainty remains high. The best example is CAR T therapy. Immune cells are taken from the patient’s blood, genetically modified to target cancer cells, and injected back to the patient’s body. Thousands of patients have been treated with the first two approved CAR T drugs: Novartis’ Kymriah and Kite Pharma’s Yescarta. The outcome, however, varies greatly1 from patient to patient. Scientists and clinicians are working hard to understand why the therapy sometimes results in miracles and sometimes causes severe side effects. We are back to the guessing game.
The problem is, as the number of options grows, so does the number of unanswered questions: Personalization alone does not provide the answers. New therapies are becoming increasingly complex, often involving different types of live cells interacting with each other in the patient body. It is a system of constantly shifting balances. Immune cells, which are our body’s natural defence, can be “trained” to find and kill cancer cells. However, they can be over stimulated and release large amounts of cytokines, chemicals that can trigger a whole-body immune response and cause toxicity. On the other hand, cancer cells are constantly evolving and may learn to escape the defence system, or even trick immune cells into turning a blind eye. Besides that, other cell types around the cancer cells—making up the tumour microenvironment—play important parts too. We are only starting to glimpse the surface of the interlocking systems.
What if we can take a snapshot of the patient’s body, and zoom in into each single cell to look at its role and status? What if we can trace the evolution of cancer colonies over time, and find out which ones are likely to perish or flourish?
“Heterogeneity is one of the most important features of cancer. Not only is every patient unique, every cancer cell is unique too. This causes huge variation in treatment response. For advanced therapies, we need better tools in order to understand each patient’s response to the treatment and to interpret the variability.”
A combination of two emerging technologies can help to achieve such a level of precision: Single cell multi-omics and artificial intelligence (AI).
Single Cell Multi-Omics: From Gene to Protein, One Cell at a Time
Named as the method of the year 2013 by Nature,2 and the breakthrough of the year 2018 by Science Magazine,3 single cell analysis is a technology that provides the most comprehensive picture of what is happening in the human body. It allows us to zoom into individual cells, and analyze thousands of DNA, RNA or protein molecules inside—the building blocks of cell function.
Single cell RNA sequencing is becoming a widely used tool for researchers to better understand disease development. It is currently used in some clinical trials to monitor treatment response, to compare patients responsive or resistant to therapies, and to identify biomarkers or potential drug targets.
Single cell multi-omics, where two or more types of biomolecules are measured at the same time, are relatively new and are still restricted to leading research labs. One powerful readout in addition to DNA or RNA is protein. In our cells, RNAs carry the “message” from DNA and serve as the template to make proteins. However, proteins are the molecules that actually perform cellular functions and are most often the drug targets. Moreover, the definition of many cell types depends on protein expression knowledge derived from decades of flow cytometry data. Flow cytometry is still the most commonly used method for single cell characterization today. Therefore, measuring protein expression is crucial for connecting the wealth of information from flow cytometry to the more recent technical advances in single cell gene expression analysis. Proteogenomic analysis, for example, simultaneously measures both protein expression and RNA expression from single cells. The protein measurement provides accurate cell typing information and can conveniently be compared to existing flow cytometry data. The RNA data allows for unbiased exploratory study to look for underlying changes in cell function and status.
International consortiums have been formed to explore this new methodology. The Human Cell Atlas4 project, very much like the Human Genome Project, aims to map the healthy human body cell by cell. The LifeTime Initiative,5 on the other hand, focuses on the mechanism of human diseases, and combines multiple technologies including single cell analysis, in vitro and in vivo models, and computational models.
From Single Cell Multi-Omics to Precision Medicine: Challenges Ahead
Despite the rapid technology development, single cell multi-omics are still not at the stage where they can guide clinical decision making. There are still many serious challenges before they can change medical practice.
- Noisy data. To make measurements from a single cell, we have to extract data from a tiny amount of material, leading to noisy data. Signals too low to be detected will show up as being not expressed, resulting in loss of information and skewing of the result. Improper sample processing, such as compromised cell viability, can also introduce unwanted errors.
- Batch effect. A sample measured today may produce different results compared to an identical sample measured only hours ago, since there may be differences in sample preparation, instrument, and reagent. Different platforms or even operators can also affect the readout.
- Data analysis expertise. Often, the bottleneck of single cell multi-omics is data analysis. The process is still labour intensive and not scalable. Clinician often lack access to sufficient bioinformatics resources to analyse the huge amount of data produced by each experiment.
As a result, there is still a large gap between the single cell multi-omics data and actionable information for clinicians to make treatment decisions.
AI Is the Key for Realizing Clinical Value of Single Cell Multi-Omics
AI methods are well known for their ability to handle complex and noisy data sets and to identify patterns that are otherwise elusive to the human eye. These methods are well poised to address the issues with single cell multi-omics data. Progress has already been made to bring the technology to the clinics.
Advanced data analysis techniques today are able to handle incomplete datasets or correct for data loss. Using proteogenomics can help to reduce the noise compared to gene expression alone by providing cross-validation. A note of caution: Garbage in, garbage out. It is best to minimize bias and errors from the sample processing itself, and not overly depend on post-processing algorithms. Emerging companies are aware of the problem and take efforts to control for them. At Proteona, we ensure high level of consistency for our in-house sample analysis following strict SOPs (Standard Operating Procedures). That gives us high quality datasets that we can trust.
Batch effect is a major issue, and it can seriously compromise the ability to compare data between different patients. To solve this problem, Proteona is collaborating6 with leading computer science experts to remove batch effects from single cell multi-omics data, and to minimize bias. One of our co-founders also recently published MapCell,7 a deep learning method for automated cell annotation, which works with data across different sample sets and even different platforms. Proteona has invested heavily in automated analysis solutions for personalized medicine.
To overcome the bottle neck of data analysis resources, AI powered pipelines are able to automate what humans can do, and also perform tasks in a way that is more consistent and efficient than manual analysis. One example is cell phenotyping. Identifying cell types can be very labour intensive, costing 5 to 10 hours of manpower to annotate a single set of data. Moreover, one person may classify the same cells very differently from another, depending on the reference gene and protein data chosen. Now, using Proteona’s automatic cell annotation algorithm, we can perform the task consistently within a couple of seconds, and to great detail. For example, we can distinguish between tumour infiltrating T cells, tissue infiltrating T cells, and peripheral T cells from the same sample. AI tools also allow for constant learning and updating. We can do this by using a carefully curated and regularly updated reference database of cells using high quality data generated in-house, as opposed to relying on lists of marker genes, which most algorithms use.
It is easy to dive into the technical details and fancy algorithms, and drift away from the real objective: To provide actionable information to help doctors with decision making. We work closely with clinicians to make sure that our data meet their quality standard and that we answer the right questions. We have already seen how our analysis can be used to identify new therapeutic options for cancer patients.
The Future of Precision Medicine Lies in Ai-Assisted Single Cell Multi-Omics Analysis
Imagine that one day, every cancer patient’s tumour will undergo single cell multi-omics analysis as part of the routine screening, so the most comprehensive information can be collected on the disease status at the cellular level. New patient information will be benchmarked against a database of other patients’ cells and their treatment outcomes and AI-based algorithms will be able to suggest a ranked list of treatment that the patient is most likely to benefit from. Single cell multi-omics will also be used in follow-ups to provide detailed monitoring of patient response and potential side effect, so that timely adjustment of the therapy regime can be made. The process will be highly automated and streamlined, so that clinicians can focus on the evidence and make data-driven, objective decisions for their patients.
The future may not be so far away. [APBN]
About the Author