Biological Sequence Fingerprints Using K-Mer Frequency Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for biological sequence analysis, such as DNA and protein fingerprinting, are computationally expensive and limited in discovering unknown features that explain differences in behavior, and existing chemical molecular analysis techniques are not applicable to biological sequences.
Innovation Solution
A computer-implemented method representing biological sequences in a fingerprint data structure using a bitset, counts, strings, or continuous values, enabling machine learning and statistical methods for Structure-Activity Relationship (SAR) and Quantitative Structure-Activity Relationship (QSAR) studies, independent of sequence type.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct sequence comparison methods like BLAST are used, then identification accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent segments biological sequences into k-mers (subsequences of length k) and represents them as frequency vectors. This segmentation transforms the original sequence comparison problem into a high-dimensional vector space comparison, enabling efficient distance calculations while preserving biological similarity information.
Solution Approach 2:
The patent changes the parameter representation from raw sequence data to k-mer frequency vectors in high-dimensional space. By transforming sequences into vectors where each dimension represents the frequency of a specific k-mer, the method enables the use of efficient vector distance calculations (cosine similarity, Euclidean distance) that are computationally much cheaper than direct sequence alignment while maintaining identification accuracy.
2Adaptability or versatility
If motif and pattern matching methods are used, then classification capability is improved, but ability to discover unknown features is limited
Solution Approach 1:
The patent maps sequences from one-dimensional string space into high-dimensional vector space where each dimension corresponds to a specific k-mer frequency. This dimensional transformation enables the application of machine learning algorithms and statistical methods that can detect complex, non-obvious patterns and unknown features that traditional motif matching cannot capture.
Solution Approach 2:
The patent replaces the mechanical motif-matching approach with a statistical and machine learning-based vector comparison system. Instead of relying on pre-defined motifs and patterns, the system uses k-mer frequency vectors that capture the compositional characteristics of sequences, enabling discovery of unknown features through statistical analysis and ML algorithms.
3Ease of operation
If existing biological fingerprinting methods are used, then sequence comparison is enabled, but applicability to machine learning methods is restricted
Solution Approach 1:
The patent transforms biological sequences into numerical k-mer frequency vectors, changing the parameter representation from categorical sequence data to continuous numerical features. This parameter transformation makes the data suitable for machine learning algorithms that require numerical input, while still maintaining the ability to perform sequence comparison through vector distance metrics.
Solution Approach 2:
The patent creates a universal representation format (k-mer frequency vectors) that serves multiple functions: it enables traditional sequence comparison through vector distance calculations, supports motif and pattern analysis, and is compatible with various machine learning algorithms. This multi-functional representation resolves the limitation of existing methods that were restricted to specific comparison approaches.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In accordance with one embodiment of the invention, features of biological sequences are represented in a fingerprint that includes a bitset, and may also include counts, strings or continuous values, for the features. The fingerprint can be used with machine learning and statistical methods. This is especially advantageous for, though not limited to, drug discovery processes. The method permits Structure-Activity Relationship (SAR) and Quantitative Structure-Activity Relationship (QSAR) studies to be performed with biological sequences.