Big data need big theory too.

Area of Science:

• Multiscale modeling
• Computational biology
• Data science

Background:

• Growing reliance on big data, machine learning, and data analytics across diverse fields.
• Perception that these methods can solve most problems without traditional scientific inquiry.
• Ease of digitized data acquisition fuels interest in data-driven approaches.

Purpose of the Study:

• Critique the limitations of pure big data approaches in science, particularly biology and medicine.
• Highlight the need for conceptual understanding beyond curve-fitting.
• Advocate for theory-guided experimental design and funding for fundamental process elucidation.

Main Methods:

• Analysis of big data and machine learning limitations in complex systems.
• Focus on weaknesses in providing conceptual accounts and handling out-of-range data.
• Emphasis on the role of theory in guiding data collection and model building.

Main Results:

• Pure big data methods often fail to provide conceptual understanding.
• Sophisticated methods like artificial neural nets primarily fit existing data.
• Data-driven approaches require vast datasets and can fail outside training data ranges.
• These methods lack inherent modeling of underlying system structures.

Conclusions:

• Theory is vital for efficient data collection, reliable predictive models, and conceptual knowledge.
• Blind big data projects with large budgets are less effective than theory-guided research.
• Increased funding is needed for understanding multiscale and stochastic processes in complex systems.