Advanced pattern recognition and batch similarity analysis using machine learning algorithms.
ChromaAnalyzer's clustering analysis combines dimensionality reduction with unsupervised machine learning to identify natural groupings in your chromatographic data. This powerful feature enables process similarity assessment, batch relationship identification, and quality pattern recognition across your manufacturing runs.
Before clustering, ChromaAnalyzer reduces the dimensionality of your metric space to reveal the most important patterns and relationships.
PCA identifies the directions of maximum variance in your transitional analysis metrics. The first two principal components capture the most significant patterns in column performance, while explained variance ratios quantify how much information each component represents. Ideal for understanding which metric combinations drive the most variation in your process.
UMAP preserves both local and global structure in your data, often revealing clusters that PCA might miss. It excels at maintaining neighborhood relationships between similar batches while clearly separating distinct process conditions. Particularly effective for identifying gradual process drift and subtle pattern changes over time.
Once the data is reduced to two dimensions, ChromaAnalyzer applies clustering algorithms to identify natural groupings in your process data.
K-means partitions your batches into spherical clusters by minimizing within-cluster variance. The algorithm automatically determines the optimal number of clusters (2-10) using silhouette analysis, which measures how well-separated and cohesive each cluster is. Excellent for identifying distinct operating regimes or process conditions.
Hierarchical clustering builds a tree of relationships between batches, using Ward linkage to minimize within-cluster sum of squares. This method reveals the hierarchical structure of your process similarity, from broad process families down to specific operational conditions. Particularly useful for understanding process evolution and identifying related batch groups.
Core Metrics: Uses the 4 key metrics (Direct AF, TransWidth, Non-Gaussian HETP, Asymmetry Factor) identified by Kim et al. as most critical for chromatography column integrity assessment in production-scale downstream processing. These metrics provide the essential indicators of column performance while reducing noise from correlated measurements.
All Metrics: Includes all 9 available transitional analysis metrics for comprehensive pattern analysis, useful when investigating subtle process variations or validating core metric findings.
Kim N, Kwon S, Kim Y, Kim G, Kim Y, Saxena L. Predictive Algorithm Modeling for Early Assessments in Downstream Processing: Using Direct Transition and Moment Analysis To Assess Chromatography Column Integrity at Production Scale. BioProcess International. 2023;March 21. Available at: https://www.bioprocessintl.com/chromatography/predictive-algorithm-modeling-for-early-assessments-in-downstream-processing-using-direct-transition-and-moment-analysis-to-assess-chromatography-column-integrity-at-production-scale
