Illumina has filed a patent for a computer-implemented method that uses machine learning models to predict probe intensity values in microarrays. The method involves receiving sample-specific image data, identifying probe intensity values and probe sequences/features, and training a machine learning model to determine predicted probe intensity values based on the input. The training data is derived from the same sample-specific image data used for testing the model. GlobalData’s report on Illumina gives a 360-degree view of the company including its patenting strategy. Buy the report here.
According to GlobalData’s company profile on Illumina, microfluidics automation was a key innovation area identified from patents. Illumina's grant share as of September 2023 was 40%. Grant share is based on the ratio of number of grants to total number of patents.
Training machine learning models to predict probe intensity values
A recently filed patent (Publication Number: US20230316054A1) describes a computer-implemented method for analyzing sample-specific image data in the context of a microarray probe. The method involves receiving the image data, which includes a signal associated with a sample for a probe in a microarray. The observed probe intensity value for the sample is identified based on the image data. Additionally, the method identifies either the probe sequence or one or more probe features that affect the total probe intensity value for the sample.
The patent further describes training a machine learning model using training data derived from the sample-specific image data. This model is used to determine a predicted probe intensity value based on inputting either the probe sequence or the probe features. The predicted probe intensity value represents the predicted total signal intensity of the sample's signal for the probe. Importantly, the training data used for the machine learning model is derived from the same sample-specific image data that may be separated for testing the trained model.
The sample-specific image data mentioned in the patent includes a raw x signal and a raw y signal, which represent fluorescent labels for genotypes A and B, respectively. The observed probe intensity value can be either a raw probe intensity value or a normalized probe intensity value, while the predicted probe intensity value can be a predicted raw probe intensity value or a predicted normalized probe intensity value.
The patent also discusses different machine learning models that can be used, such as linear regression models, random forest models, and neural networks. The input for these models can include k-mer features of the probe, an entire predefined set of probe features, or the probe sequence itself. In the case of a neural network, a hybrid neural network with a convolutional portion and a fully-connected feed forward portion is mentioned.
The system described in the patent includes an imaging system for capturing image data and generating sample-specific image data. The system also includes at least one processor for receiving the image data, identifying the observed probe intensity value, identifying the probe sequence or probe features, and training the machine learning model.
Overall, this patent presents a computer-implemented method and system for analyzing sample-specific image data in the context of microarray probes. The method involves training a machine learning model to predict probe intensity values based on inputting probe sequence or probe features. The system includes an imaging system and a processor for implementing the method.