# US 11,908,549 B2 · Bayesian inference > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,908,549 B2 - Document ID: US11908549B2 - Status: Granted - Kind: B2 - Application number: 16/646,187 - Filed: 2018-09-21 - Published: 2020-08-27 - Granted: 2024-02-20 - Earliest claimed: 2017-09-28 - Assignees: Koninklijke Philips N.V. - Applicants: Koninklijke Philips N.V. - Inventors: Martijn Theodorus Lambert Akse; Wilhelmus Franciscus Johannes Verhaegh - Web page: https://patentgrep.com/patent/US11908549B2 ## CPC - G16B 5/20: Probabilistic models - G16B 40/20: Supervised data analysis - G16B 40/00: ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding - G16B: BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY ## Abstract The invention relates to a computer-implemented Bayesian inference method for performing Bayesian inference in a Bayesian network, which includes a continuous child node (3) and its discrete parent node (2) having two states. Being provided with a calibrated continuous probability distribution of the observations of the child node (3) for each of the two states of the parent node (2), the inferring for a new observation of the child node (3) for which the state of the parent node (2) is not known of the probability of at least a first state of the parent node (2) makes use of masses of portions of the probability distributions that depend on a locational relationship of the probability distributions and the value of the new observation. This may ensure that the inferred probability of the first state of the parent node (2) monotonically changes with monotonically changing values of the new observation. ## Claims ### Claim 1 A computer-implemented Bayesian inference method for performing Bayesian inference in a Bayesian network, which includes a continuous child node and discrete parent node of the child node, the discrete parent node having two states, wherein the computer-implemented Bayesian inference method comprises: a) providing continuous probability distributions of observations of the child node fore the two states of the parent node, respectively, based on values of observations of the child node for which the states of the parent node are known, wherein the probability distributions comprise a lower-located probability distribution at a relatively lower location with respect to the values of the observations of the child node and a higher-located probability distribution at a relatively higher location with respect to the values of the observations of the child node; and b) inferring for a new observation of the child node, for which the state of the parent node is not known, a probability of at least a first state of the two states of the parent node, wherein the inferring includes calculating (i) for the lower-located probability distribution, a mass of a portion of the lower-located probability distribution that is larger than a value of the new observation of the child node and (ii) for the higher-located probability distribution, a mass of a portion of the richer-located probability distribution that is smaller than the value of the new observation of the child node, and calculating the probability of the first state based on the calculated masses. ### Claim 2 (depends on 1) The computer-implemented Bayesian inference method of in claim 1, wherein the calculating of the masses includes determining which of the probability distributions is the lower-located probability distribution and which is the higher-located probability distribution based on a location parameter of the probability distributions. ### Claim 3 (depends on 2) The computer-implemented Bayesian inference method of claim 2, wherein the location parameter of the probability distributions is a mean value of a median value. ### Claim 4 (depends on 1) The computer-implemented Bayesian inference method of claim 1, wherein the calculating of the probability of the first state includes normalizing one of the calculated masses by the sum of the calculated masses. ### Claim 5 (depends on 1) The computer-implemented Bayesian inference method as of claim 1, wherein the inferring of the probability of the first state further includes transforming the calculated probability of the first state, wherein the transforming depends on a measure of similarity of the probability distributions. ### Claim 6 (depends on 5) The computer-implemented Bayesian inference method of claim 5, wherein the transforming includes scaling the calculated probability of the first state and applying an offset to the scaled calculated probability of the first state. ### Claim 7 (depends on 6) The computer-implemented Bayesian inference method of claim 6, wherein the scaling includes performing a comparison of the probability distributions or of the observations of the child node on which the probability distributions are based to determine the measure of similarity, and scaling the calculated probability of the first state in dependence of a function of the determined measure of similarity. ### Claim 8 (depends on 7) The computer-implemented Bayesian inference method of claim 7, wherein the function of the determined measure of similarity is a linear function. ### Claim 9 (depends on 7) The computer-implemented Bayesian inference method of claim 7, wherein the function of the determined measure of similarity is a non-linear function. ### Claim 10 (depends on 9) The computer-implemented Bayesian inference method of claim 9, wherein the non-linear function comprises a Hill function. ### Claim 11 (depends on 1) The computer-implemented Bayesian inference method of claim 1, wherein the parent node has further states, wherein the continuous probability distribution of the observations of the child node is provided for each of the states of the parent node, wherein the masses are calculated for successive pairs of the states of the parent node and the probability of the first state is calculated based on the calculated masses. ### Claim 12 (depends on 11) The computer-implemented Bayesian inference method of claim 11, wherein the calculating of the probability of the first state includes calculating odds between successive states of the parent node based on the calculated masses for the states. ### Claim 13 (depends on 12) The computer-implemented Bayesian inference method of claim 12, wherein the probability of the first state is calculated based on a product of the calculated odds. ### Claim 14 A Bayesian inference device for performing Bayesian inference in a Bayesian network, which includes a continuous child node and a discrete parent node of the child node, the discrete parent node having two states, wherein the Bayesian inference device comprises: a computer; and a non-transitory storage medium storing instructions that, when executed by the computer, cause the computer to: provide continuous probability distributions of observations of the child node for the two states of the parent node, respectively, based on values of observations of the child node for which the states of the parent node are known, wherein the probability distributions comprise a lower-located probability distribution at a relatively lower location with respect to the values of the observations of the child node and a higher-located probability distribution at a relatively higher location with respect to the values of the observations of the child node; and infer for a new observation of the child node, for which the state of the parent node is not known, a probability of at least a first state of the two states of the parent node, wherein the inferring includes calculating (i) for the lower-located probability distribution, a mass of a portion of the lower-located probability distribution that is larger than a value of the new observation of the child node and (ii) for the higher-located probability distribution, a mass of a portion of the richer-located probability distribution that is smaller than the value of the new observation of the child node, and calculating the probability of the first state based on the calculated masses. ### Claim 15 (depends on 14) The Bayesian inference device of claim 14, wherein the instructions cause the computer to calculate the probability of the first state by normalizing one of the calculated masses by the sum of the calculated masses. ### Claim 16 (depends on 14) The Bayesian inference device of claim 14, wherein the instructions cause the computer to infer of the probability of the first state by further transforming the calculated probability of the first state, wherein the transforming depends on a measure of similarity of the probability distributions, and includes scaling the calculated probability of the first state and applying an offset to the scaled calculated probability of the first state. ### Claim 17 (depends on 1) A method for inferring activity of a cellular signaling pathway using probabilistic modelling of target gene expressions, wherein the method comprises: inferring the activity of the cellular signaling pathway in a medical subject based at least on expression levels of one or more target genes of the cellular signaling pathway measured in an extracted sample of the medical subject, wherein the inferring comprises: determining an activity level of a transcription factor element in the extracted sample of the medical subject, the transcription factor element controlling transcription of the one or more target genes of the cellular signaling pathway, the determining being based at least in part on evaluating a Bayesian network according to the method of claim 1 relating expression levels of the one or more target genes of the cellular signaling pathway to the activity level of the transcription factor element; and inferring the activity of the cellular signaling pathway in the medical subject based on the determined activity level of the transcription factor element in the extracted sample of the medical subject. ### Claim 18 (depends on 17) The method of as defined in claim 17, further comprising: determining whether the cellular signaling pathway is operating abnormally in the medical subject based on the inferred activity of the cellular signaling pathway in the medical subject. ### Claim 19 (depends on 17) The method of claim 17, further comprising: recommending prescribing a drug for the medical subject that corrects for abnormal operation of the cellular signaling pathway, wherein the recommending is performed when the cellular signaling pathway is determined to be operating abnormally in the medical subject based on the inferred activity of the cellular signaling pathway. ### Claim 20 A non-transitory computer readable medium storing instructions for performing Bayesian inference in a Bayesian network, which includes a continuous child node and a discrete parent node of the child node, the discrete parent node having two states, wherein when executed by a computer, the instructions cause the computer to: provide continuous probability distributions of observations of the child node for the two states of the parent node, respectively, based on values of observations of the child node for which the states of the parent node are known, wherein the probability distributions comprise a lower-located probability distribution at a relatively lower location with respect to the values of the observations of the child node and a higher-located probability distribution at a relatively higher location with respect to the values of the observations of the child node; and infer for a new observation of the child node, for which the state of the parent node is not known, a probability of at least a first state of the two states of the parent node, wherein the probability of at least the first state is inferred by calculating (i) a mass of a portion of the lower-located probability distribution that is larger than a value of the new observation of the child node and (ii) a mass of a portion of the higher-located probability distribution that is smaller than the value of the new observation of the child node, and calculating the probability of the first state based on the calculated masses.