INVESTIGADOR(ES) PRINCIPAL(ES):

Mauricio Alexander Alvarez Lopez

4 horas

CODIGO CIE

6-14-3

Horas

TIPO DE CONVOCATORIA

2013. British Council

TIPO DE PROYECTO

Investigación Aplicada

OBJETIVO(S)

Objetivo General del Proyecto: Desarrollar un conjunto de algoritmos basados en inferencia Bayesiana de tipo ralo para realizar la selección del modelo de fuerza latente en una red de regulación de genes de múltiples factores de transcripción. Objetivos específicos: 1) Desarrollar un método de inferencia Bayesiana que incluya funciones de probabilidad a priori de tipo ralas para ser aplicado en selección del modelo de fuerza latente en una red de regulación de genes con múltiples factores de transcripción. 2) Validar los algoritmos para la selección del modelo de fuerza latente ralo para la ingeniería inversa de múltiples factores de transcripción en diferentes redes de regulación de genes.

RESUMEN

Latent force models (LFM) are a hybrid approach, which combines multiple output Gaussian processes and differential equations, where the covariance functions encode the physical models given by the differential equations. LFM require the specification of the number of latent functions used to build the covariance function for the outputs. Furthermore, they assume that the output data is explained by using the entire set of latent functions, which is not the case in many real applications. We propose the use of an Indian Buffet process (IBP) as a way to perform model selection over the number of latent Gaussian processes in LFM applications. Furthermore, IBP allows us to infer the interconnections between latent functions and the outputs. Latent force models (LFM) are a hybrid approach of Gaussian processes (GP), where the covariance function is built from a convolution. This convolution is performed using the Green's function of a differential equation. Hence, latent functions may represent a physical quantity, like the action of a protein for transcription regulation of a gene or a latent force in a system involving masses, springs and dampers. Despite its success for prediction, it is still unclear how to select the number of latent functions and, how to unveil the interactions between the latent functions and the output variables that are being modeled. We have introduced a new variational method to perform model selection in latent force models. Our main aim was to identify the relationship between the latent functions and the outputs in LFM applications. The proposed method achieved comparable results to the DTCVAR method, in which, a full connectivity between latent functions and output functions is assumed. This makes our method suitable to applications where the complexity of the model should be reduced.

ESTADO

Concluido

FECHA DE INICIO

02/04/2014

FECHA DE FINALIZACION

02/04/2016

PRODUCTOS

Gaussian Processes Summer School

Curso diseñado para programas de maestría

IBPLFM: Indian Buffet Process for Latent Force Models

Software

URL

Indian Buffet process for model selection in convolved multiple-output gaussian processes

Artículo publicado en Revista de divulgación

URL

Indian Buffet Process for Model Selection in Latent Force Models

Ponencia en evento especializado

Optimización Bayesiana

Curso diseñado para programas de maestría

Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications

Artículos en revista A1 ó A2