Fig 1.
BRAF modeling flowchart: From a biological question to validated personalized logical models.
Fig 2.
Logical modeling principles and personalization.
(A) A logical model with three nodes: the regulatory graph, the corresponding logical rules and the transition rates as used in MaBoSS [31]. (B) Part of the state transition graph with the two possible transitions resulting from the given initial conditions and the probabilities of choosing stochastically one of them. (C) Schematic representation of a logical model simulation with MaBoSS: average trajectory obtained from the mean of many individual stochastic trajectories. (D) Personalization with discrete data (e.g., mutations) with some nodes forced to 0 based on loss of function alteration (left) or 1 based on gain of function/constitutive activation (right). (E) Personalization with continuous data used to define the initial conditions of nodes and to influence the transitions rates and the subsequent probabilities of transition in asynchronous update; the graph on the left represents the normalized values of genes A, B and C for patients 1, 2 and N; the right side represents the personalization of logical model using values from patient N (red profile), first defining the initial probabilities of node activation (middle) and then influencing the probabilities of transitions from one state to another (right): here, since gene A is highly expressed in the red patient, the probability of activation of the corresponding node is increased (resp. probability of inhibition is decreased for gene B).
Fig 3.
Cell lines data: Mutations and sensitivities to BRAF inhibition.
(A) Distribution of the assigned mutations in the four most frequently mutated genes in the colorectal/melanoma cohort of cell lines [37]. (B) Differential sensitivities to BRAF inhibition by the drug PLX-4720 (upper panel) or by CRISPR inhibition (lower panel), depending on BRAF mutational status and cancer type. Numbers of cell lines in each category are indicated. Note that high sensitivities correspond to low AUC and high scaled Bayesian factors.
Fig 4.
Logical model of signaling pathways around BRAF in colorectal and melanoma cancers.
Grey nodes represent input nodes, which may correspond to the environmental conditions. Blue nodes accounts for families. Light blue node represents the output of the model. Square nodes represent multi-valued variable (MEK, ERK, p70 and Proliferation). Note that Proliferation is used as the phenotypic read-out of the model.
Table 1.
List of assertions used to validate the logical model.
Fig 5.
Validation of personalized models with cell lines data.
(A) Pearson correlations between normalized Proliferation scores from personalized models and experimental sensitivities to BRAF inhibition by drug or CRISPR targeting; each row corresponds to a different personalization strategy; only the values for the significant correlations are displayed. (B) Scatter plots with non-overlapping points corresponding to correlations of panel A, with the three personalization strategies, focusing one one drug (PLX-4720) and one CRISPR dataset (Broad) only. (C) Enlargement of the scatter plot comparing model scores (personalized with mutations and RNA) and experimental sensitivity to CRISPR targeting of BRAF (left) with the corresponding table representing the omics profiles used for each cell line to explore the response mechanisms. This panel can be advantageously replaced by one of the interactive plots proposed in the provided code.
Fig 6.
Application of personalized models to other CRISPR targets.
(A) Personalization strategies using either mutations only (as discrete data) or combined with RNA (as continuous data) with their corresponding scatter plots in panels B and C. (B) Scatter plot comparing normalized Proliferation scores of p53 inhibition in the models with experiment sensitivity of cell lines to TP53 CRISPR inhibition, indicating p53 mutational status as interpreted in the model. Pearson correlations and the corresponding p-values are shown. (C) Similar analysis as in panel B with PI3K model node and PIK3CA CRISPR inhibition.