Calls for Papers (special): International Journal of Quantitative Structure-Property Relationships (IJQSPR)

Special Issue On: Application of Machine Learning Theories in QSAR/QSPR

Submission Due Date

Guest Editors
Giuseppina Gini, DEIB, Politecnico di Milano, Italy

Many forces drive modern QSAR/QSPR models:
- first of all the requirements of creating new chemicals with wanted properties in fields as large as drugs, cosmetics, biocides, and industrial products;
- second, the growing concern about the risks of chemicals that has produces advanced norms to regulate their use;
- third, the availability of more data in the public domain that opened the door to more advanced models;
- finally, the tremendous improving of hardware and software that allows more complex modeling, as represented in particular by the development of Artificial Intelligence methods.
About 20 years ago the Artificial Intelligence (AI) and the Toxicology Communities joined together in the so-called “Predictive Challenge”, where a few tens of molecules were given to researchers to produce a QSAR model of cancerogenicity. The results then showed that it is possible to build a QSAR model only using the chemical structure. In the following years QSAR methods started to embrace new learning methods, to incorporate more kinds of descriptors, and to move from linear to non-linear models.
More recently, a new challenge promoted by the pharm industry was launched in 2012 with thousands of chemical structures; among the many models developed a deep neural net was found to be at the core of the winning model.

Even though Artificial Intelligence methods are commonly used in modeling physical and biological properties of chemicals, there is a lack of clearly assessing their advantages or disadvantages in building QSAR/QSPR. Moreover, AI models may incorporate both symbolic and implicit knowledge representations, and extracting and organizing knowledge from such models is still challenging. This special issue will try to give answer to questions such as:

• How AI based methods work in making accurate and predictive QSAR/QSPR models?
• What kind of knowledge is represented or hidden in AI-based models?
• Which kind of chemical and biological knowledge should and could be given to AI based models?
• How and why users can accept AI-based models?
• What are the next challenges for QSAR/QSPR methods?

Recommended Topics
• Application of AI tools, including Support Vector Machines and Neural Networks, to QSAR/QSPR
• Use of ensemble methods, as Random Forests, to build QSAR/QSPR
• Learning methods for modeling: pros and cons
• Deep Neural Nets and deep learning for building QSAR/QSPR
• Learning from chemical structures
• AI methods to define and choose descriptors
• AI methods to automatically extract relevant functional subgroups to build SAR models
• AI methods to integrate SAR and QSAR
• Extracting knowledge from AI models of chemical/biological properties
• Interpreting AI models in terms of chemical and biological knowledge
• AI methods to integrate statistical results and expert knowledge
• AI tools to integrate in vivo and in vitro data
• Acceptance of AI methods in various user contexts
• Hardware and software to develop AI-based QSAR/QSPR models
• Theoretical developments of new QSARs using machine learning principles

Submission Procedure
Researchers and practitioners are invited to submit papers for this special theme issue on Application of machine learning theories in QSAR/QSPR on or before November 15, 2018. All submissions must be original and may not be under review by another publication. INTERESTED AUTHORS SHOULD CONSULT THE JOURNAL’S GUIDELINES FOR MANUSCRIPT SUBMISSIONS at All submitted papers will be reviewed on a double-blind, peer review basis. Papers must follow APA style for reference citations.

All submissions and inquiries should be directed to the attention of:
Giuseppina Gini
Guest Editor
International Journal of Quantitative Structure-Property Relationships (IJQSPR)