Back to results
US 2024/0404636 A1Published application
Oligonucleotide-based machine learning
AssigneeCreyon Bio, Inc.
InventorsSwagatam MUKHOPADHYAY, Christopher E. HART
Life of this patent
Estimating the term…
Abstract
A machine-learned model can be trained on and applied to oligonucleotide data. The machine-learned model can be, for example, a neural network, a random forest classifier, or a regression model, and can be trained in one or more stages. The machine-learned model can be applied in design settings, for instance by being configured to predict biophysical effects corresponding to oligonucleotides, by processing real-world experimental or laboratory data, and by retraining the machine-learned model in response to the processed data.
Claims30
A method for training a machine-learned model, comprising:
- generating a first set of oligonucleotides by performing all possible single n-gram mutations of an initial oligonucleotide to obtain oligonucleotides mapped to a distributed range of measures of having a biophysical effect such that each n-gram of the initial oligonucleotide is represented at least once by the first set of oligonucleotides;
- synthesizing a subset of the first set of oligonucleotides;
- administering the synthesized subset of the first set of oligonucleotides to one or more cells;
- measuring, for each oligonucleotide of the subset of the first set of oligonucleotides, the biophysical effect on the one or more cells;
- creating a first training set based, at least in part, on the synthesized subset of the first set of oligonucleotides and the measures of the biophysical effect;
- training a machine-learned model in a first stage using the first training set, the machine-learned model configured to map an oligonucleotide sequence to a measure of the biophysical effect;
- after training the machine-learned model in the first stage, generating a second set of oligonucleotides such that each n-gram of the first set of oligonucleotides is represented at least once by the second set of oligonucleotides, the second set of oligonucleotides selected based on an output of the machine-learned model such that the second set of oligonucleotides corresponds to a distributed range of measures of the biophysical effect, the second set of oligonucleotides including an approximately equal number of oligonucleotides that correspond to the biophysical effect and that do not correspond to the biophysical effect;
- creating a second training set comprising, for each oligonucleotide of the second set of oligonucleotides, a sequence of the oligonucleotide and a measure of the biophysical effect of the oligonucleotide determined based on real-world experimental determination;
- and
- training the machine-learned model in a second stage using the second training set.
7 more elements
A method for training a machine-learned model, comprising:
- generating a first set of oligonucleotides by performing all possible single n-gram mutations of an initial oligonucleotide to obtain oligonucleotides mapped to a distributed range of measures of having a biophysical effect such that each n-gram of the initial oligonucleotide is represented at least once by the first set of oligonucleotides;
- creating a first training set comprising, for each oligonucleotide of the first set of oligonucleotides, a sequence of the oligonucleotide and a measure of the biophysical effect of the oligonucleotide determined based on real-world experimental determination;
- training a machine-learned model in a first stage using the first training set, the machine-learned model configured to map an oligonucleotide sequence to a measure of the biophysical effect;
- after training the machine-learned model in the first stage, generating a second set of oligonucleotides such that each n-gram of the first set of oligonucleotides is represented at least once by the second set of oligonucleotides, the second set of oligonucleotides selected based on an output of the machine-learned model such that the second set of oligonucleotides corresponds to a distributed range of measures of the biophysical effect, the second set of oligonucleotides including an approximately equal number of oligonucleotides that correspond to the biophysical effect and that do not correspond to the biophysical effect;
- synthesizing a subset of the second set of oligonucleotides;
- administering the synthesized subset of the second set of oligonucleotides to one or more cells;
- measuring, for each oligonucleotide of the subset of the second set of oligonucleotides, the biophysical effect on the one or more cells;
- creating a second training set based, at least in part, on the synthesized subset of the second set of oligonucleotides and the measures of the biophysical effect;
- and
- training the machine-learned model in a second stage using the second training set.
7 more elements
A method for training a machine-learned model, comprising:
- generating a first set of oligonucleotides by performing all possible single n-gram mutations of an initial oligonucleotide to obtain oligonucleotides mapped to a distributed range of measures of having a biophysical effect such that each n-gram of the initial oligonucleotide is represented at least once by the first set of oligonucleotides;
- creating a first training set comprising, for each oligonucleotide of the first set of oligonucleotides, a sequence of the oligonucleotide and a measure of the biophysical effect of the oligonucleotide determined based on real-world experimental determination;
- training a machine-learned model in a first stage using the first training set, the machine-learned model configured to map an oligonucleotide sequence to a measure of the biophysical effect;
- after training the machine-learned model in the first stage, generating a second set of oligonucleotides such that each n-gram of the first set of oligonucleotides is represented at least once by the second set of oligonucleotides, the second set of oligonucleotides selected based on an output of the machine-learned model such that the second set of oligonucleotides corresponds to a distributed range of measures of the biophysical effect, the second set of oligonucleotides including an approximately equal number of oligonucleotides that correspond to the biophysical effect and that do not correspond to the biophysical effect;
- creating a second training set comprising, for each oligonucleotide of the second set of oligonucleotides, a sequence of the oligonucleotide and a measure of the biophysical effect of the oligonucleotide determined based on real-world experimental determination;
- and
- training the machine-learned model in a second stage using the second training set.
4 more elements
Dependencies are inferred from the claim text.