Systematic error detection in the database of liquid crystals (LiqCryst) using predictive models.

Area of Science:

• Materials Science
• Data Science
• Chemistry

Background:

• Experimental data frequently contain anomalies, which can stem from errors or represent unexplored scientific frontiers.
• Identifying and classifying these anomalies is crucial for ensuring data reliability and fostering new discoveries.
• Machine learning (ML) offers powerful tools for anomaly detection in large datasets.

Purpose of the Study:

• To develop and validate a human-in-the-loop ML approach for scrutinizing data integrity in materials science.
• To leverage domain expertise alongside ML for accurate anomaly identification in liquid crystal phase transition data.
• To differentiate between data errors and potentially novel scientific insights.

Main Methods:

• An ML model was trained on the comprehensive LiqCryst 5.2 database of liquid crystalline (LC) material phase transition behaviors.
• A human-in-the-loop system was implemented, integrating ML anomaly detection with expert human review.
• Anomalies detected by the ML model were systematically re-examined by domain experts to ascertain their origin (error vs. new knowledge).

Main Results:

• The ML model successfully identified multiple anomalies within the LiqCryst 5.2 database, a widely accepted industry standard.
• Human experts confirmed the ML model's effectiveness in detecting inconsistencies in reported LC phase transition data.
• The study demonstrated the model's capability to flag anomalies that warrant further investigation for potential scientific breakthroughs.

Conclusions:

• The proposed human-in-the-loop ML methodology effectively enhances the scrutiny of experimental data integrity.
• This approach can reliably detect errors and highlight unexplored phenomena, thereby stimulating future scientific discoveries.
• The adaptable methodology holds broad applicability across diverse materials systems and scientific domains for data validation.