According to Gartner, “Through 2022, 85% of AI projects will deliver erroneous outcomes due to bias in data, algorithms, or the teams responsible for managing them.”
That might sound like bad news for computational biology, where AI is integral to the predictive modeling process. But the key to proving the Gartner prediction wrong, assuming the data and algorithms are correct, lies in the project management.
Timing Is [Almost] Everything
Timing is an important factor in a computational biology project. You also need the right people involved – which usually is not a problem since each member of the active team brings the requisite expertise.
But here’s where timing comes in: Computational biology projects are like sprint relays. Members of the team work together, but in stages. And knowing the exchange zone – when to “pass the baton” to another expert and when to “accept the baton” – is crucial to a project’s success, particularly when AI is involved.
Here, we examine how to manage your computational biology hand-offs with athletic precision – for the best project outcomes at the finish line.
Stage 1: Strategy and Preparation
As with any research, computational projects are designed to help answer an important scientific question or solve a complex problem, such as identifying the best target or drug candidate.
Because a computer model only predicts in a computational sense, a biological perspective must be involved early in the project scope to ensure that what the model is designed to predict is the correct outcome. Experts cannot substitute their knowledge for others’, or the result will reflect that bias. When designing a computational project—and throughout the stages of the project—it’s important to incorporate both verification and validation:
- Verification of a model confirms that it operates correctly from a computational perspective, maintaining high accuracy and producing the right results for the input-output relationships.
- Validation, typically biological validation, determines whether the model is showing the right relationship from a scientific perspective. The model must not only work with the inputs given, but it also must solve the right problem to be useful and accurate.
In the project definition stage, then, it is critical to have both the computer scientists and the biologists involved. Each team member contributes unique strengths – expertise and perspective – to confirm that the right questions are being asked in order to ensure that they are aligned with the anticipated solution.
Stage 2: Defining the Exchange Zones
It can be difficult to determine early in the projects exactly where hand-offs need to happen. During the project, defining “exchange zones” for these hand-offs to occur enables team members to quickly access data, launch the computational tools they need, and reassemble different project components.
Exchange zones also serve another critical function – they can be tracked as milestones that indicate the project is on-track toward the expected solution.
A loose schedule of exchanges should be defined up front, keeping in mind that if those hand-offs are too early, a team member’s efforts might be wasted; if hand-offs are too late, they can put the project in jeopardy of delays and missed deadlines.
Stage 3: The Hand-Off
Once the project is defined, just as in a relay race, members of the team perform specific subtasks of the larger project. Some of these subtasks are carried out simultaneously without any interaction with other team members. Other subtasks require hand-offs in the predefined (Stage 2) exchange zones.
As an example, a subtask for an experimental scientist could be to generate data. Once that data has been generated, the experimental scientist must pass the data to a computational scientist to analyze. The computational scientist’s subtask is to run the model using that generated data. The experimentalist will then pick up the baton again and verify the analysis and experimental results are aligned with the goals of the experiment.
Exchange zones are critical for two reasons:
- They serve as Agile-like checkpoints along the way to the project finish line – and allow for project course-correction before it is too late.
- They bring together the right team members – the biologist, for example – for critical validation that the model is predicting as expected and that it is answering the right questions. Remember, the model predicts in a computational sense, but if the inputs are incorrect, the outcome will be as well.
Stage 4: The Finish Line
Once a project gets to the finish line, if timed well, all of the exchanges along the way should have provided the critical validation that the model is sound and solves the right problem.
It is at this final stage that the model must be officially validated. Recalling the Gartner prediction – a project will fail if it cannot be operationalized, if it solves the wrong problem, or if there is bias in data or algorithm. But the underlying issue – AI or not – is if the project isn’t a collaborative team effort, it will not succeed.
The success of computational projects begins with a solid strategy that defines the project and ensures that the problem or question you are trying to solve is the right one.
Setting up exchange zones along the course of the project provide opportunities for course-correction. They also ensure a non-biased, multi-disciplinary view of the project that verifies – at critical points in the project process – that the problem the model is trying to solve is the right one. The smoother that process is the faster the team delivers and achieves great discoveries.
Finally, the outcome requires an official validation of project processes, data sets, and modeling. AI cannot replace human reasoning and expertise. Models will predict, but unless the inputs are correct, the outcomes will fail. In the end, computational project success is defined by whether the model prediction is valid for the market it is designed to serve.
Want to learn more? Read our blog, 9 Challenges Computational Researchers Need to Address in 2022.