Systems and methods for quantifying multiscale competitive landscapes of clonal diversity in glioblastoma
Methods that implement image-guided tissue analysis, MRI-based computational modeling, and imaging informatics to analyze the diversity and dynamics of molecularly-distinct subpopulations and the evolving competitive landscapes in human glioblastoma multiforme (“GBM”) are provided. Machine learning models are constructed based on multiparametric MRI data and molecular data (e.g., CNV, exome, gene expression). Models can also be built based on specific biological factors, such as sex and age. Inputting MRI data into the trained predictive models generates maps that depict spatial patterns of molecular markers, which can be used to quantify and co-localize regions molecularly distinct subpopulations in tumors and other regions, such as the non-enhancing parenchyma, or brain around tumor (“BAT”) regions.
1. A method for constructing and implementing a machine learning model to generate at least one image that depicts spatial patterns of a molecular marker across a region-of-interest in a subject, the steps of the method comprising:
constructing a trained machine learning model by:
(i) accessing training data with a computer system, the training data comprising magnetic resonance imaging (MRI) data acquired from one or more subjects and molecular data determined from biopsies collected from the one or more subjects, wherein the molecular data comprises DNA copy number variation (CNV) data and exome data;
(ii) quantifying regional molecular diversity in the one or more subjects from the molecular data;
(iii) training a machine learning model based on the training data and the quantified regional molecular diversity in the one or more subjects, wherein the machine learning model is trained on the training data to localize molecularly distinct subpopulations and phenotypic niches across a region-of-interest; and
generating an image that depicts spatial patterns of a molecular marker across a region-of-interest in a subject by inputting magnetic resonance images acquired from the subject to the trained machine learning model.
2. The method as recited in claim 1 , wherein the region-of-interest includes a brain around tumor (BAT) region.
3. The method as recited in claim 1 , wherein the training data further comprises biological factor data including at least one of a sex or an age of each of the one or more subjects.
4. The method as recited in claim 3 , wherein the biological factor data includes the sex of each of the one or more subjects and step (iii) includes training a first machine learning model based the training data and the quantified regional molecular diversity in the one or more subjects associated with a female sex and training a second machine learning model based the training data and the quantified regional molecular diversity in the one or more subjects associated with a male sex.
5. The method as recited in claim 4 , wherein the image whose pixel values quantify regional molecular diversity in the region-of-interest in the subject is generated by inputting magnetic resonance images acquired from the subject to the first machine learning model when the subject is a female and to the second machine learning model when the subject is a male.
6. The method as recited in claim 1 , wherein the molecular data corresponds to a plurality of different molecular markers and step (iii) includes training a different machine learning model for each of the plurality of different molecular markers based on the training data and the quantified regional molecular diversity in the plurality of subjects associated with each different molecular marker.
7. The method as recited in claim 6 , wherein the molecular marker includes at least one of EGFR, PDGFRA, PTEN, NF1, TP53, CDKN2A, RB1, ATRX, or MET.
8. The method as recited in claim 1 , wherein quantifying regional molecular diversity in the plurality of subjects from the molecular data includes quantifying intra-biopsy heterogeneity.
9. The method as recited in claim 8 , wherein intra-biopsy heterogeneity is quantified based on a presence of a molecular marker throughout biopsy sub-regions, an absence of a molecular marker throughout biopsy sub-regions, or an admixture of a molecular marker being present and absent throughout biopsy sub-regions.
10. The method as recited in claim 1 , wherein quantifying regional molecular diversity in the one or more subjects from the molecular data includes quantifying correlations between different molecular markers.
11. The method as recited in claim 1 , wherein the molecular data further comprises at least one of RNA sequencing (RNA-seq), gene expression, proteomics analysis data, or combinations thereof.
12. The method as recited in claim 11 , wherein quantifying regional molecular diversity in the one or more subjects from the molecular data includes classifying molecularly distinct subpopulations based on at least one of the CNV data and the exome data.
13. The method as recited in claim 11 , wherein the molecular marker is a phenotypic marker and quantifying regional molecular diversity in the one or more subjects from the molecular data includes quantifying phenotypic niches based on RNA-seq.
14. The method as recited in claim 13 , wherein the phenotypic marker indicates at least one of hypoxia, cell proliferation, angiogenesis, cell invasion, or combinations thereof.
15. The method as recited in claim 1 , wherein the MRI data comprises magnetic resonance images and parametric maps generated from the magnetic resonance images.
16. The method as recited in claim 15 , wherein the magnetic resonance images include at least one of T1-weighted images, post-contrast T1-weighted images, T2-weighted images, post-contrast T2-weighted images, T2*-weighted images, post-contrast T2*- weighted images, diffusion-weighted images, perfusion-weighted images, or combinations thereof.
17. The method as recited in claim 15 , wherein the parametric maps include at least one of relative cerebral blood volume, mean diffusivity, fractional anisotropy, or combinations thereof.
18. The method as recited in claim 15 , wherein the MRI data further comprises texture feature maps generated from at least one of the magnetic resonance images or the parametric maps.
19. The method as recited in claim 18 , wherein the texture feature maps include texture features computed based on at least one of a gray-level co-occurrence matrix (GLCM), local binary patterns (LBP), a discrete orthonormal Stockwell transform (DOST), or a Gabor filter.
20. The method as recited in claim 1 , wherein the trained machine learning model is a transfer learning model trained on MRI data and molecular data acquired from the subject.
21. The method as recited in claim 1 , wherein quantifying regional molecular diversity in the one or more subjects from the molecular data comprises quantifying the molecular diversity based on one or more continuous variables.