Current visualization of reads from a 1000bp sequence:

t-SNE

UMAP

PCA

Develop a function to benchmark the proposed embedding approach to true positions of reads.

1. Fix the order of reads, calculate the relative distance (measure the variation of the distance)
2. Sort the distance, calculate the maximum order difference (avoid FP)

Discussion with Chatgpt for the development workflow.

Steps:

1. K-mer Representation:

   • Convert each DNA read into a k-mer frequency vector.

2. Dimensionality Reduction (Optional):

   • Use PCA or another technique to reduce the dimensions of your k-mer vectors.

3. Grid-based Partitioning:

   • Divide the (reduced) feature space into grids and assign each k-mer vector to a grid cell.

4. Local Graph Construction:

   • Within each grid cell, construct a graph where nodes represent individual reads.
   • Create edges between nodes based on the actual overlap of the reads.

5. Edge Weights (Optional):

   • Assign weights to the edges based on the degree of overlap.

6. Global Graph Integration:

   • Connect graphs from different grid cells based on overlapping reads that fall into adjacent cells.

7. Graph Refinement:

   • Remove or merge nodes/edges based on specific criteria, like low overlap or edge weight.

8. Analysis:

   • Apply graph analytics to identify patterns, like connected components that might indicate bacteria strains.

Potential Issues:

1. Scalability: Creating a graph within each grid cell can still be computationally expensive for large datasets.

2. Boundary Cases: Reads near the edge of one grid cell might actually be more related to reads in an adjacent cell, leading to missed connections.

3. Parameter Tuning: The size of the grid and the criteria for edge creation need to be carefully selected.

Adding grid-based partitioning as a step can help manage computational costs by allowing you to work on smaller subsets of the data at a time. Would you like to explore any of these steps in more detail?

Visualize simulated read alignments, compare the alignments with embedding difference: may further develop into a benchmarking framework.

Perform k-mer representation, PCA and grid partitioning to extract local data points. The purpose of this step is to build a local graph based on sequencing overlapping using local data points.