JPWO2021034712A5

JPWO2021034712A5 -

Info

Publication number: JPWO2021034712A5
Application number: JP2022509680A
Authority: JP
Publication date: 2023-08-18

Claims

1. A computer-implemented method for detecting dysregulation in cellular pathways in patient sample transcriptomic data, comprising:
(a) for one or more trained pathway disruption engines comprising a plurality of trained machine learning models, including identifiers and expression levels for one or more genes in said transcriptome data; preparing the transcriptome data in the form of a transcriptome value set, comprising:
The plurality of trained machine learning models comprising:
1) at least one pathway-level machine learning model trained to identify dysregulation in an input transcriptome value set associated with a cellular pathway;
2) at least one module-level machine learning model trained to identify dysregulation in the input transcriptome value set associated with the module;
3) at least one gene-level machine learning model trained to identify dysregulation in the input transcriptome value set associated with the gene; and 4) associated with variants or associated with variants of unknown significance. comprising at least one variant-level machine learning model trained to identify dysregulation in the input transcriptome value set;
(b) selecting a plurality of trained machine learning models from at least one path breaking engine to apply to said transcriptome value set;
(c) applying a selected plurality of trained machine learning models to said transcriptome value set to generate at least one pathway disruption score indicative of dysregulation in a cellular pathway.

Each pathway level trained machine learning model is trained based on training data comprising a plurality of positive control specimens and a plurality of negative control specimens, each positive control specimen comprising genetic data and a positive control the genetic data comprises at least one detectable pathogenic variant in at least one gene comprised within the cellular pathway, each negative control specimen comprising genetic data, the negative control genetic data comprised within the cellular pathway does not contain a pathogenic variant in any of the genes
each module-level trained machine learning model is trained based on training data comprising a plurality of positive control specimens and a plurality of negative control specimens, each positive control specimen comprising genetic data; the genetic data comprises at least one detectable pathogenic variant in at least one gene contained within the module, each negative control specimen comprising the genetic data, and the negative control genetic data contained within the module; does not contain a pathogenic variant in any of the genes of
Each gene-level trained machine learning model is trained based on training data comprising a plurality of positive control samples and a plurality of negative control samples, each positive control sample comprising genetic data, a positive control the genetic data contains at least one detectable pathogenic variant in the gene, each negative control specimen contains genetic data, and the negative control genetic data contains no pathogenic variant in the gene;
each variant-level trained machine learning model is trained on training data comprising a plurality of positive control specimens and a plurality of negative control specimens, each positive control specimen comprising genetic data; the genetic data contains the variant, each negative control specimen contains genetic data, the negative control genetic data does not contain the pathogenic variant;
The method of Claim 1.

at least one trained pathway-level model, at least one trained module-level model, at least one trained gene-level model, or at least one trained variant-level model combined with multiple positive controls trained on training data containing the specimen and multiple negative control specimens,
at least one of said training comprises calculating a plurality of differential metric values between positive control specimens and negative control specimens, each differential metric value being included in said pathway or said module; and selecting genes with differential metric scores at or below a threshold for inclusion in the model. , the method of claim 2.

3. The method of claim 2, wherein at least a portion of the positive control gene data and the negative control gene data comprises DNA data.

3. The method of claim 2, wherein at least a portion of the positive control gene data and the negative control gene data comprises RNA data.

6. The method of claim 5, wherein the RNA data comprises transcriptome data.

6. The method of claim 5, wherein detectable pathogenic variants comprise RNA expression levels.

6. The method of claim 5, wherein the negative control RNA transcriptome data does not contain significant variation in expression levels for expressed RNA compared to one or more wild-type samples.

2. The method of claim 1, comprising generating a path disruption report based on at least one path disruption score, and presenting the path disruption report to at least one of a display or memory.

The pathway disruption report includes at least one pathway disruption score and associated information, wherein the information is a) the likely causative mutation, b) the identification of one or more variants of unknown significance, c) one or multiple recommended therapies, d) proposals for monitoring organoids after exposure to treatment based on pathway disruption scores, e) subject and associated patients based on pathway disruption scores, at least one clinical trial and d) including at least one of the reference medical literature.

receiving a first pathway disruption score indicative of cellular pathway dysregulation in a cellular pathway from a first trained pathway disruption engine;
receiving a second pathway disruption score indicative of cellular pathway dysregulation in a cellular pathway from a second trained pathway disruption engine;
2. The method of claim 1, comprising generating a metapathway depiction based on the cellular pathway, the first pathway disruption score, and the second pathway disruption score, and presenting the metapathway representation on a display.

at least one trained path disruption engine including a trained model configured to output a disruption model score, wherein a model score below a predetermined threshold indicates dysregulation and a model above the predetermined threshold 2. The method of claim 1, wherein the score indicates dysregulation.

at least one trained pathway disruption engine comprising a plurality of trained models, a cellular pathway comprising one or more gene modules, and each trained model comprised in the plurality of trained models , configured to output model scores associated with different genetic modules contained within a cellular pathway.

14. The method of claim 13, further comprising calculating a global dysregulation score based on model scores output by each of the trained models.

2. The method of claim 1, wherein the plurality of machine learning models comprises one or more linear regression machine learning algorithms or non-linear models.

2. The method of claim 1, wherein the cellular pathway comprises 1-5 genes, 6-10 genes, 10-20 genes, or 20-100 genes.

generating a pathway disruption report comprising a depiction of a cellular pathway including a number of modules contained within the cellular pathway and an indication of dysregulation in at least one of the modules contained within the cellular pathway; 2. The method of claim 1, comprising presenting the report on at least one of a display or memory.

2. The method of claim 1, wherein the cellular pathway is a rat sarcoma/receptor tyrosine kinase (RAS/RTK) pathway level machine learning model.

At least one trained path breaking engine has (a), (b) or both (a) and (b):
(a) module-level machine learning models for at least the following modules: RAS module, RAF module, MEK module and ERK module,
The RAS model is trained to identify dysregulation in the input transcriptome value set associated with alterations in at least one of the KRAS, NRAS and HRAS genes, and the RAF model is trained to identify at least one of the RAF1, BRAF and ARAF genes. A MEK model was trained to identify dysregulation in the input transcriptome value set associated with a change in the MAP2K1 gene, and a MEK model was trained to identify dysregulation in the input transcriptome value set associated with a change in the MAP2K1 gene; an ERK model trained to identify dysregulation in an input transcriptome value set associated with alterations in at least one of the MAPK3 and MAPK1 genes;
Each trained model outputs a model score to a trained pathway disruption engine, which produces a pathway disruption score indicative of dysregulation in cellular pathways;
(b) the at least one trained pathway disruption engine comprises a gene-level machine learning model of at least one of the following genes: KRAS, NRAS, HRAS, RAF1, BRAF, ARAF, MAP2K1, MAPK3 and MAPK1;
19. Each trained model outputs a model score to a trained pathway disruption engine, wherein the trained pathway disruption engine produces a pathway disruption score indicative of dysregulation in cellular pathways. The method described in .

2. The method of claim 1, wherein the at least one trained pathway disruption engine comprises a phosphatidylinositol-3-kinase (PI3K) pathway-level machine learning model.

At least one trained path breaking engine has (a), (b) or both (a) and (b):
(a) a module-level machine learning model for at least one of the following modules: PI3K module, AKT1 module, TOR module and PTEN module;
The PI3K model was trained to detect dysregulation in the input transcriptome value set associated with alterations in at least one of the PIK3CA and PICKCB genes, and the AKT1 model was associated with alterations in at least one of the AKT1, AKT2 and AKT3 genes. and the TOR model is trained to detect dysregulation in the input transcriptome value set associated with alterations in at least one of the RICTOR, RPTOR and MTOR genes. a PTEN model trained to detect dysregulation of an input transcriptome value set associated with alterations in at least one of the PTEN, PIK3R1, PIK3R2 and PIK3R3 genes;
Each trained model outputs a model score to a trained pathway disruption engine, which produces a pathway disruption score indicative of dysregulation in cellular pathways;
(b) a gene-level machine learning model for at least one of the following genes: PIK3CA, PICKCB, AKT1, AKT2, AKT3, RICTOR, RPTOR, MTOR, PTEN, PIK3R1, PIK3R2 and PIK3R3,
21. The method of claim 20, wherein each trained model outputs a model score to a trained pathway disruption engine, the pathway disruption engine generating a pathway disruption score indicative of dysregulation in cellular pathways. Method.

2. The method of claim 1, wherein multiple models are selected by the user.

2. The method of claim 1, wherein at least one trained path breaking engine comprises a custom path-level machine learning model.

selecting one or more detectable pathogenic variant positive controls to be used to train a pathway level model, optionally wherein each of the selected positive controls is more than one detectable and/or the step of selecting comprises filtering the positive and negative controls for clinical characteristics, and/or the positive control detectable pathogenic variants comprise one or more insertions, 3. The method of claim 2, comprising deletions, rearrangements, copy number alterations or substitutions.