Synex Biosystems
Predictive modelling for experimental decisions
Synex builds predictive models for a specific experimental objective. We work with data from your process to estimate an outcome and help you decide what to do next. A model does not replace lab judgement. It is not universal software either. It gets defined, evaluated and validated for the decision you need to make.
What we do
Predicting an outcome means estimating, from the measurements you already have, what might happen to a sample, a condition or a material. Using that prediction means turning it into an actual decision. What to prioritise. When to move forward. Where you need one more measurement. Before any of that, we agree on what a useful prediction looks like for your workflow.
Problems we can explore
Prioritising samples or conditions
Work out which material deserves the next assay when there isn't time to test everything.
Anticipating readiness for the next step
Estimate whether a sample meets the conditions to move forward in the process.
Identifying what else to measure
Point to where one more check could cut down an important uncertainty.
Illustrative example
A laboratory has some initial measurements and historical processing outcomes. We look at whether that data can anticipate which new samples will meet later requirements. This example doesn't represent a real client or an observed result. It's only here to show the kind of decision that can be studied.
Input data
We can work with assay results, instrument measurements, molecular data, images and process records. To connect them we usually need to identify the samples, know which outcome we're predicting and know when each measurement was taken. Batches, conditions, centres and instruments can matter too, along with permission to use the data. Images aren't always needed. Not every project needs every kind of data either. Having a lot of data doesn't guarantee it's useful. There's no sample count that works for every problem. Sometimes more data needs to be collected.
How we work
01 · Define
Agree the decision, outcome and criteria that make the prediction useful.
02 · Assess
Review the available data, how it relates to the objective and whether the work is feasible.
03 · Develop and compare
Build and compare models against simple alternatives and the current procedure where possible.
04 · Validate
Test behaviour in relevant conditions and look at which errors matter to the lab.
05 · Prepare for use
Agree how predictions get obtained, interpreted and worked into the process.
What the delivery may include
Scope gets agreed per project. It can include a feasibility assessment, a model built for the agreed goal, a report covering errors and limitations, and an agreed way to get and use the predictions. It can also define what's needed for integration and further validation. We don't assume there's already a standard platform, API, dashboard or integration in place. The delivery format depends on your process and your team.
Validation and limits
When the case calls for it, we separate donors, batches, centres or time periods so we evaluate on data that wasn't used for training. As inputs, we only use the measurements actually available at the moment of the decision. We compare against simple alternatives or the current procedure and look at which errors actually matter. A retrospective result isn't the same as validating something in real use. A model also can't be assumed to transfer to another lab without testing it there. A feasibility assessment can end up concluding there isn't enough basis for a usable model.
Frequently asked questions
Do we need images?
Not necessarily. Images are one possible source among others. What makes sense to assess depends on the objective and the data you have.
Can you work with our historical results?
Yes. We can assess whether they're useful when they relate to a defined decision and outcome.
Do our data need to be organised?
We need to identify samples, measurements and outcomes. We can assess where things stand and point out what further preparation is needed.
How many samples do we need?
There is no universal number. It depends on the objective, the variability and how the model gets validated.
Does the model replace assays?
No. It can help guide priorities or checks. Decisions about assays stay part of the lab's own workflow.
What do we receive?
The scope may include a model, its evaluation, its limitations and an agreed way to use it. Formats are defined per project.
How do we start?
Tell us what decision you want to anticipate, what data you have and when you need to decide. We talk through fit before defining a project.