Explaining Symbolic Regression Predictions

Filho, Renato Miranda; Lacerda, Anísio; Pappa, Gisele L.

doi:10.1109/cec48606.2020.9185683

Cited by 20 publications

(10 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prior ML work in this area has attained similar performances, though it lacks the explainability of symbolic regression (Filho, Lacerda, & Pappa, 2020; Khalid, Tuszynski, Szlek, Jachowicz, & Mendyk, 2015). Utilizing a combination of random forest, the least absolute shrinkage and selection operator (LASSO) and logistic regression, Xu et al (2017) obtained a sensitivity of 83.3% and a specificity of 90.5% on their validation set.…”

Section: Discussionmentioning

confidence: 99%

A Symbolic Regression Approach to Hepatocellular Carcinoma Diagnosis Using Hypermethylated CpG Islands in Circulating Cell-Free DNA

Goyal¹

2022

Preprint

View full text Add to dashboard Cite

Hepatocellular carcinoma is the most common primary liver cancer, accounting for 90% of cases, and a major cause of death worldwide. Despite this, alpha-fetoprotein tests are the only blood-based diagnostic tools available, and their use is limited by their low sensitivity. DNA methylation changes, which have been implicated in a majority of cancers, offer an alternative method of diagnosis through measuring such changes in circulating cell-free DNA present in blood plasma. A genetic programming-based symbolic regression approach was applied to gain the benefits of machine learning while avoiding the opacity drawbacks of "black box" models. The data included plasma samples from 36 patients with hepatocellular carcinoma as well as a control group of 55 that contained patients with and without cirrhosis. A 75-25 train-test splitting was done before training. The symbolic regression methodology developed an equation utilizing the methylation levels of three biomarkers, with an accuracy of 91.3%, a sensitivity of 100%, and a specificity of 87.5% on the test data. All three biomarkers are differentially methylated in cancerous and non-cancerous samples. The performance matches prior research while providing the added benefits of transparency. Circulating cell-free DNA presents opportunities for minimally invasive early diagnosis of hepatocellular carcinoma, and utilizing transparent machine learning approaches like symbolic regression can allow accurate diagnosis by combining biological and mathematical principles. Future validation of the model obtained here on a larger and more diverse dataset can reveal the potential for such approaches in cancer diagnosis and open the way for further research.

show abstract

Section: Discussionmentioning

confidence: 99%

A Symbolic Regression Approach to Hepatocellular Carcinoma Diagnosis Using Hypermethylated CpG Islands in Circulating Cell-Free DNA

Goyal¹

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Based on the expanded dataset, we employ two symbolic regression methods, namely brute-force linear regression and genetic programming. 94,95 To further reduce difficulties, we decompose the rates for nucleation and growth as:…”

Section: Development Of the Negen1 Modelmentioning

confidence: 99%

“…This allows us to fit the sub-datasets separately with fewer parameters corresponding to those in the individual terms. Brute-force linear regression was performed based on a function database of ~1 million, while genetic programming 94,95 was used multiple times and the best term was kept (having simultaneously good accuracy and simplicity). The complete form of the rate expression can be obtained by:…”

Section: Development Of the Negen1 Modelmentioning

confidence: 99%

Machine-Learning Discovered Crystallization Model for Two Dimensional Covalent Organic Frameworks: Towards Precise Control of the Crystal Quality

Tian,

2024

Preprint

View full text Add to dashboard Cite

The rational molecular design and experimental condition optimizations for two-dimensional co-valent organic frameworks (2D COFs) call for a crystallization model capable of capturing exper-imental time and size scales. However, accurately describing their crystallization process remains a significant challenge due to the presence of non-classical pathways. Here, we demonstrate the implementation of a machine-learning approach, overcoming the difficulties associated with bot-tom-up model derivation. The resulting model, referred to as NEgen1, establishes correlations among the induction time, nucleation rate, growth rate, material parameters, and common solu-tion synthesis conditions for 2D COFs that belong to the nucleation-elongation category. NEg-en1 represents the emergence of practical crystallization models for 2D COFs, enabling the direct calculation of their crystallization processes in both experimental times and sizes. The results elu-cidate the detailed competition between the nucleation and growth dynamics in solution, which has been inappropriately apprehended via classical, empirical models with assumptions invalid for 2D COFs. Importantly, we demonstrate the potential application of the NEgen1 model in opti-mizing the synthesis conditions, which has predominantly relied on empirical knowledge to date. The identification of conditions superior to those routinely used experimentally reveals a promis-ing strategy of gradually increasing monomer addition speed for growing large 2D COF crystals while maintaining a reasonable synthesis time. These results highlight the potential for systemati-cally improving the crystal quality of 2D COFs for wider applications.

show abstract

“…To address this issue, we can use an even simpler model to explain the complex GP model. For example, Filho et al [250] proposed to use linear model to approximate the local predictions of a complex GP model around a given datapoint for symbolic regression. The local dataset contains a number of nearest neighbours from the training set (rather than randomly sampling).…”

Section: B Post-hoc Local Interpretability By Gpmentioning

confidence: 99%

Explainable Artificial Intelligence by Genetic Programming: A Survey

Mei

Chen

Lensen

et al. 2023

IEEE Trans. Evol. Computat.

View full text Add to dashboard Cite

Explainable artificial intelligence has received great interest in the recent decade, due to its importance in critical application domains such as self-driving cars, law and healthcare. Genetic programming is a powerful evolutionary algorithm for machine learning. Compared with other standard machine learning models such as neural networks, the models evolved by GP tend to be more interpretable due to their model structure with symbolic components. However, interpretability has not been explicitly considered in genetic programming until recently, following the surge in popularity of explainable artificial intelligence. This paper provides a comprehensive review of the studies on genetic programming that can potentially improve the model interpretability, both explicitly and implicitly, as a byproduct. We group the existing studies related to explainable artificial intelligence by genetic programming into two categories. The first category considers the intrinsic interpretability, aiming to directly evolve more interpretable (and effective) models by genetic programming. The second category focuses on posthoc interpretability, which uses genetic programming to explain other black-box machine learning models, or explain the models evolved by genetic programming by simpler models such as linear models. This comprehensive survey demonstrates the strong potential of genetic programming for improving the interpretability of machine learning models and balancing the complex trade-off between model accuracy and interpretability.

show abstract

Explaining Symbolic Regression Predictions

Cited by 20 publications

References 13 publications

A Symbolic Regression Approach to Hepatocellular Carcinoma Diagnosis Using Hypermethylated CpG Islands in Circulating Cell-Free DNA

A Symbolic Regression Approach to Hepatocellular Carcinoma Diagnosis Using Hypermethylated CpG Islands in Circulating Cell-Free DNA

Machine-Learning Discovered Crystallization Model for Two Dimensional Covalent Organic Frameworks: Towards Precise Control of the Crystal Quality

Explainable Artificial Intelligence by Genetic Programming: A Survey

Contact Info

Product

Resources

About