Objectives
Last updated
Last updated
You can set your objectives based on the assays that you defined. You must select one primary objective and can add multiple constraint objectives.
A primary objective is the main optimization goal for your project. You can choose to either maximize or minimize its value by selecting "Increase" or "Decrease." Only one primary objective can be set per round.
A constraint objective is a target that the models aim to achieve while optimizing for the primary objective. We recommend setting at most four constraint objectives.
Our models currently require one primary objective, and you can choose to add single, multiple, or no-constraint objectives. Each assay can be associated with only one primary objective or constraint objective.
Note: As your project evolves, objectives or their priority may change. Thus, objectives are not fixed and can be updated at the beginning of each round.
The primary objective dictates whether the associated property in the project should increase or decrease.
To set the primary objective:
Select the assay for the property for which you would like to set a primary objective.
Specify whether the property should increase or decrease.
Constraint objectives are targets that the models aim to achieve while optimizing for the primary objective.
To set a constraint objective:
Select the assay for the property for which you would like to set a constraint.
Determine whether the property should remain above or below the threshold.
Set a threshold. This can be your project's base sequence, a sequence from your uploaded data, or an absolute value.
If you set an absolute value as the threshold, the models will generate sequences with property performance above or below that value, depending on your specification. Alternatively, if you select a sequence as the threshold, the models will generate sequences with property performance above or below the property of the selected sequence.
A generated sequence will "satisfy" a constraint if it meets or exceeds the target value or performs as specified relative to a reference sequence. Our machine learning platform computes a probability that a given sequence will satisfy all constraints. It tries to maximize this probability while simultaneously optimizing the primary objective.
Note: This simultaneous optimization does not guarantee that we can achieve your constraints in a single round. We simply try to increase the probability that the sequences satisfy it incrementally each round.
Once your objectives are set, you can start with Data benchmark if you want to evaluate whether your project's data is suitable for Cradle's ML or proceed directly to Sequence generation.
