Drivers Predicting the Future: Using Driver Mutations to Estimate Tumor Growth Patterns

Image courtesy of Pixabay.

Ever since that split-second in which your parents’ gametes fused to generate your life, the cells that make up who you are have not stopped dividing, even right now. As they replicate, it is likely that their DNA will mutate, incorporating new traits into the genetic code that writes their fate––what structure they will adopt, what function they will perform, what purpose they will serve. While some mutations can be responsible for traits like ginger hair or the absence of wisdom teeth, or even at times go completely unnoticed, other far more dangerous ones hold the power to corrupt a cell’s machinery, culminating in grave ripple effects that can make all systems go haywire. 

The word “tumor” is laced with terrifying potential. Possibilities of anarchical growth, silent spread, and rapid lethality render these neoplasms’ behavior difficult to predict. In an effort to circumvent this uncertainty, a group of Yale researchers led by Dr. Mark Gerstein, Albert L. Williams Professor of Biomedical Informatics and Professor of Molecular Biophysics & Biochemistry, Computer Science, and Statistics & Data Science, published a paper in February reporting a mathematical model they developed. The model looks at a specific kind of mutation called driver mutations to estimate a tumor’s growth pattern. 

Driver Mutations

The development of cancer is an evolutionary process, punctuated by mutations. Historically, several theories have been put forward as to how researchers can study such genetic alterations, but, most recently, mutations have been increasingly labeled as either drivers or passengers to categorize them according to their relevance in tumor progression. Leonidas Salichos, a postdoctoral associate and first author of the paper, explained that “we have a lot of mutations in every tumor, sometimes thousands of mutations, and a few of them are what we call drivers, … which we try to detect because those are the ones that actually play a role in tumor progression.” Conversely, mutations identified as non-significant in terms of tumor development are dubbed passenger mutations. 

According to Gerstein, driver mutations can be defined as the “few mutations that accumulate in the cell and drive its growth forward.” In the paper, the authors discuss different means through which these mutations can trigger the formation of tumors, including hindering the ability of tumor-suppressor genes from impeding tumor growth and enhancing the level of expression of oncogenes, which are genes that can cause cancer. 

How the Model Works

When tumors are biopsied, a sample is often extracted and sequenced to reveal its genetic composition. According to Salichos, the number of times a specific position in the genome is sequenced is very important. The deeper the sequencing, the more accurate you can expect the measurement of a mutation’s frequency within a population to be. At the end of the process, you have acquired a run-down that details all of the mutations detected as well as their respective frequencies, which paints a clear picture of their expressivity within the cell’s genetic code. 

“Based on the frequency, you can already make an assessment of whether that mutation happened early or late in the tumor, because, if it happened early in the tumor progression, we are expecting it to have a higher frequency at the end,” Salichos said. Therefore, ordering mutations from those that appear most to least provides insight into the order in which they occurred. This information contextualizes what mutations might have stimulated tumor growth and which ones occurred as collateral damage, helping frame their relevance with respect to tumor progression. 

Salichos explained that, based on this idea, he developed a mathematical model that uses the frequency of some of the mutations that happened exactly before the driver mutation to detect presence of the driver and estimate tumor growth at the precise moment when it first emerged. This examination allowed the group to gauge the impact of this phenomenon, since the introduction of a driver mutation into a sample often creates a detectable perturbation in the variant allele frequency distribution.  “Once you introduce a driver into a population that grows, now the population starts growing faster, and that has an impact on the frequency of the mutations that happen before that,” Salichos said. 

Driving Towards More Accurate Cancer Prognosis

“Traditionally, the way people find these drivers is they look at cohorts of cancer patients at the same time,” Gerstein said. Salichos also highlighted that over a thousand samples are often needed in traditional methods, since large numbers are required to ground observations that something deviates from the normal. At least computationally, this is how scientists normally validate a suspicion that a specific mutation is important for the development of a tumor. 

However, the need to examine a whole cohort can serve as a limitation in the study of cancer genomics. If several samples are required every time physicians want to understand the role of a driver in a tumor’s progression within a particular patient, individualized assessments of how specific growths will develop become more complicated to attain. In that regard, this is where their model adds something new. “This method doesn’t require a cohort, but only one tumor to be very deeply sequenced,” Gerstein explained. The approach incorporates ultra-deep sequencing, a method that entails the sequencing of the same location in the genome several times to identify rare variations, into their analysis. “The novelty of this method was, instead of looking into many different samples, we actually harnessed the frequency of the mutations based on growth models and analyzed both the mutations and their frequency in the population to try to make an assessment of which of them mattered and which did not, all within that individual sample,” Salichos said. 

This model could enable scientists to account for how cancer heterogeneity results in no two tumors ever being completely alike. While reliance upon averages is often important when looking at growths that behave differently depending on their genetic make-up, as well as the context in which they are inserted, every tumor––even ones of the same kind––will behave differently. “With this kind of model, you can look into an individual’s tumor in a more direct way… you don’t have to think about a cohort or a database very much,” Gerstein noted. Considering how this framework’s applicability does not require more than a single tumor, it could lay the foundation for more specialized evaluations that take only the characteristics of the studied tumor into account, making more specific assessments possible. 

Testing the model’s efficacy

In order to test the model’s effectiveness, simulations were run to see if it could, in fact, predict the presence, time of occurrence, and effect of a driver mutation. In addition to testing the algorithmic function upon which the model relied by applying it under different growth models, such as exponential growth and logistic growth, the group also sought to demonstrate the framework’s efficacy on real samples. To that end, the model was applied to 993 tumors obtained from the Pan-Cancer Analysis of Whole Genomes Consortium––an online database that provides information obtained through whole genome sequencing and integrative analysis data of over 2,600 tumors across 38 diverse types of tumor. 

After observing that the identified drivers were correlated with periods of positive growth in the samples examined, the group sought to further consolidate their framework by applying it to a sample of an Acute Myeloid Leukemia (AML) tumor. According to Gerstein, this tumor was chosen due to its history of having been deeply sequenced in the past. For AML, the growth patterns they predicted showed significant similarities with those exhibited by the tumor.

The promising evidence surrounding the model’s effectiveness provides reasons to be optimistic about its future applications. This novel way to look into tumors could make a big difference in the future of cancer treatments. Instead of just relying on broad data, this could allow doctors to tailor their evaluations of a patient’s prognosis to what their specific tumor sample shows. In this way, this kind of personalized assessment could herald a new era in cancer genomics.

About the Author: Maria Fernanda Pacheco is a first-year student and prospective Neuroscience and Comparative Literature double-major in Grace Hopper college. In addition to writing and editing for the YSM, Maria writes and edits for the Yale Global Health Review, is a staff reporter for the Yale Daily News SciTech desk, acts as a volunteer teacher for Community Health Educators in Mental Health workshops and participates in Yale’s chapter of Timmy Global Health.

Acknowledgements: The author would like to thank Dr. Leonidas Salichos and Prof. Mark Gerstein for their time, availability and enthusiasm in sharing their research.


How Many Of These Common Genetic Mutations Do You Possess?. (2020). Retrieved 16 March 2020, from

Mirny, L. (2020). Driver and Passenger Mutation in Cancer. Retrieved 20 March 2020, from

Deep Sequencing. (2020). Retrieved 12 March 2020, from

Oncogene. (2020). Retrieved 9 March 2020, from