A Bayesian data fusion based approach for learning genome-wide transcriptional regulatory networks

Sauta, Elisabetta; Demartini, Andrea; Vitali, Francesca; Riva, Alberto; Bellazzi, Riccardo

doi:10.1186/s12859-020-3510-1

“…Since popularization by Pearl (1986), learning Bayesian Networks (BNs) has solidified into a steadfast research area for 40 years. It has become an important paradigm for modeling and reasoning under uncertainty and has seen applications from stock market prediction (Malagrino, Roman, and Monteiro 2018) and medical diagnosis (Shih, Choi, and Darwiche 2018) to Gene Regulatory Networks (GRNs) (Sauta et al 2020). Despite Bayesian Network Structure Learning (BNSL) being NP-hard (Chickering, Heckerman, and Meek 2004) and even simpler structures like polytrees being NPhard(er) (Dasgupta 1999), new constraints (Grüttemeier and Komusiewicz 2020), improvements (Trösser, de Givry, and Katsirelos 2021), and scalings (Scanagatta et al 2015) are presented at major AI conferences every year.…”

Section: Introductionmentioning

confidence: 99%

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Yakaboski¹,

Santos²

2023

Preprint

0

View full text Add to dashboard Cite

Successful machine learning methods require a trade-off between memorization and generalization. Too much memorization and the model cannot generalize to unobserved examples. Too much over-generalization and we risk under-fitting the data. While we commonly measure their performance through cross validation and accuracy metrics, how should these algorithms cope in domains that are extremely under-determined where accuracy is always unsatisfactory? We present a novel probabilistic graphical model structure learning approach that can learn, generalize and explain in these elusive domains by operating at the random variable instantiation level. Using Minimum Description Length (MDL) analysis, we propose a new decomposition of the learning problem over all training exemplars, fusing together minimal entropy inferences to construct a final knowledge base. By leveraging Bayesian Knowledge Bases (BKBs), a framework that operates at the instantiation level and inherently subsumes Bayesian Networks (BNs), we develop both a theoretical MDL score and associated structure learning algorithm that demonstrates significant improvements over learned BNs on 40 benchmark datasets. Further, our algorithm incorporates recent off-the-shelf DAG learning techniques enabling tractable results even on large problems. We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA).

show abstract

“…Since popularization by Pearl (1986), learning Bayesian Networks (BNs) has solidified into a steadfast research area for 40 years. It has become an important paradigm for modeling and reasoning under uncertainty and has seen applications from stock market prediction (Malagrino, Roman, and Monteiro 2018) and medical diagnosis (Shih, Choi, and Darwiche 2018) to Gene Regulatory Networks (GRNs) (Sauta et al 2020). Despite Bayesian Network Structure Learning (BNSL) being NP-hard (Chickering, Heckerman, and Meek 2004) and even simpler structures like polytrees being NPhard(er) (Dasgupta 1999), new constraints (Grüttemeier and Komusiewicz 2020), improvements (Trösser, de Givry, and Copyright © 2023, Association for the Advancement of Artificial Intelligence (www.aaai.org).…”

Section: Introductionmentioning

confidence: 99%

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Yakaboski

¹

,

Santos

²

2023

AAAI

1

0

View full text Add to dashboard Cite

Successful machine learning methods require a trade-off between memorization and generalization. Too much memorization and the model cannot generalize to unobserved examples. Too much over-generalization and we risk under-fitting the data. While we commonly measure their performance through cross validation and accuracy metrics, how should these algorithms cope in domains that are extremely under-determined where accuracy is always unsatisfactory? We present a novel probabilistic graphical model structure learning approach that can learn, generalize and explain in these elusive domains by operating at the random variable instantiation level. Using Minimum Description Length (MDL) analysis, we propose a new decomposition of the learning problem over all training exemplars, fusing together minimal entropy inferences to construct a final knowledge base. By leveraging Bayesian Knowledge Bases (BKBs), a framework that operates at the instantiation level and inherently subsumes Bayesian Networks (BNs), we develop both a theoretical MDL score and associated structure learning algorithm that demonstrates significant improvements over learned BNs on 40 benchmark datasets. Further, our algorithm incorporates recent off-the-shelf DAG learning techniques enabling tractable results even on large problems. We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA).

show abstract

“…Obviously, it is a key issue to effectively fuse the multi-sensor information. Many techniques are proposed to solve the issue, such as the Dempster-Shafer theory (DST) [1,2], Kalman filtering (KF) [3,4], fuzzy theory [5], Bayesian reasoning method [6,7], neural network [8], and so on. However, there are many uncertainties, for example, randomness and imprecision, in the real world.…”

Section: Introductionmentioning

confidence: 99%

A New Total Uncertainty Measure from A Perspective of Maximum Entropy Requirement

Huang

¹

,

Deng

²

,

Jiang

³

2021

Entropy

View full text Add to dashboard Cite

The Dempster-Shafer theory (DST) is an information fusion framework and widely used in many fields. However, the uncertainty measure of a basic probability assignment (BPA) is still an open issue in DST. There are many methods to quantify the uncertainty of BPAs. However, the existing methods have some limitations. In this paper, a new total uncertainty measure from a perspective of maximum entropy requirement is proposed. The proposed method can measure both dissonance and non-specificity in BPA, which includes two components. The first component is consistent with Yager’s dissonance measure. The second component is the non-specificity measurement with different functions. We also prove the desirable properties of the proposed method. Besides, numerical examples and applications are provided to illustrate the effectiveness of the proposed total uncertainty measure.

show abstract

A Bayesian data fusion based approach for learning genome-wide transcriptional regulatory networks

Cited by 5 publications

References 96 publications

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

A New Total Uncertainty Measure from A Perspective of Maximum Entropy Requirement

Contact Info

Product

Resources

About