Logos Analytics
Logos Analytics is our in-house data science platform developed to help predict clinical trial outcomes.
Biotechnology is a data driven sector. Clinical trial outcomes dictate value creation or destruction, particularly for small and mid-sized companies. The role of data science at Logos is to simulate clinical data for a given trial and apply statistical tests to replicate what the company and FDA will do to determine if the trial was statistically significant and clinically meaningful (or not). There is no formulaic or algorithmic approach to definitively state that a trial will succeed or fail. Each investment presents its own unique project, data and frameworks to consider.
The inputs we use are varied based on available data. The analyses are dependent on the trial design as well as our assumptions and questions. However, the output is effectively the same – a refined probability of success that allows Logos to better understand the risk and the likelihood of reward. Logos Analytics tracks clinical data, builds comprehensive event calendar maps, and sources inputs from a wide variety of academic literature and drug development programs. Although we primarily use Logos Analytics to help determine probability of success or failure in clinical trials, we sometimes share our analysis with management teams.
Data Aggregation
- [Types of Data] The variety of data used in each analysis is dependent on the specific investment question. For example, underwriting a pivotal Phase 3 trial requires gathering clinical data, while earlier stage investments may require pharmacokinetic and pharmacodynamic data.
- [Data Sources] There is a wealth of relevant data, but it requires scouring a variety of potential sources, including the companies’ own trials, competitors’ trials, published patents, and academic literature.
- [Data Cleaning] All of these data need to be interpreted carefully. For example, a single-arm, investigator-sponsored clinical trial is given less weight than a double-blind, placebo control trial from a competitor. Finally, we have a variety of tools to digitize this data, to be used in subsequent analyses.
Data Aggregation
Model Validation
- All our frameworks are motivated by insights from biology and statistical clinical trial design. As much as possible, we try to make our models falsifiable: if our assumptions are true, can the model make a prediction that we can test with existing data? Models that can be externally validated are given more weight in our analysis.
- Finally, our background with our frameworks ultimately guides our decision process. For example, from hard-won experience, we know there are limitations to certain models, and we constantly perform post-mortem analyses to update those limitations.
Model Validation
Predictive Matrix
- Our analyses attempt to quantify the probability of success, and to provide confidence intervals around that probability. One general insight of this process is to highlight key parameters, where the probability of success is highly contingent on one or two things going right.
- These probabilities, of course, are weighed against the market probability of success, and we enter situations where our views are strongly divergent from the market.
