BORIS JURIČ
MACHINE LEARNING
APLICATIONS ON RNA-SEQ DATA
RNA-sequencing
• Presence and quantity

• Gene expression, transcriptome
assembly, differential analysis

• Disease biomarkers, diagnostics
• Cancer research

• Data contaminated

• High sequence similarity ~85%
PDX
Standard methods
• Xenome - kmer index table

• NGS Disambiguate - two alignments

• Pure alignments methods
• Convolutional neural network

• encoding 
Our approach
A = [1,0,0,0]
T = [0,1,0,0]
C = [0,0,1,0]
G = [0,0,0,1]
ACC…TGG

  [1,0,0,0]
  [0,0,1,0]
  [0,0,1,0]
       …
  [0,1,0,0]
  [0,0,0,1]
  [0,0,0,1]
Model assisted alignment
• Goal is gene expression quantification

• Model predicts gene, alignment only local

• Output is n-dimensional space
• Kmer content defines distance

• Similar genes closer together

• Random projection to reduce number of dimensions

• Relative distance preserved
Search space and metric
AAAAAAAA AAAAAAAT ... GGGGGGGG
DDX11L1 1 0 ... 1
WASH7P 0 1 ... 0
MIR6859-1 0 0 ... 1
MIR1302-2HG 1 0 ... 0
… … … … …
High dimensional regression
• Model predicts coordinates

• Tree search the matrix

• Alignment on N closest neighbours
So far
• Works on small amount of genes 

• Uneven gene size, insufficient metric

• Reference data structure needs improvement
Current work
• Clustering kmers from the whole genome

• Generating search space in similar fashion

• Model predicts cluster, best alignment is chosen
Thank you for attention