Basecalling for stochastic sequencing processes

• US 10,648,027 B2
• Filed: 08/04/2017
• Issued: 05/12/2020
• Est. Priority Date: 08/08/2016
• Status: Active Grant

• Current Assignee: Roche Sequencing Solutions, Inc. (Roche Holding AG)
• Original Assignee: Roche Sequencing Solutions, Inc. (Roche Holding AG)
• Inventors: Mannion, John, Mager, Morgan
• Primary Examiner(s): Leong, Susan D
• Application Number:
  US15/669,207
• Publication Number:
  US 20180037948A1
• Time in Patent Office: 1,012 Days
• CPC Class Codes: C12Q 1/6869 Methods for sequencing
• CPC Class Codes: C12Q 2535/122 Massive parallel sequencing
• CPC Class Codes: C12Q 2537/165 Mathematical modelling, e.g. logarithm, ratio
• CPC Class Codes: C12Q 2565/631 being a biochannel or pore
• CPC Class Codes: G01N 27/44791 Microapparatus sample containers with integrated microfluidic structures B01L3/5027
• CPC Class Codes: G01N 33/48721 Investigating individual macromolecules, e.g. by translocation through nanopores Coulter counters in general G01N15/12; fabrication methods for nanoscale apertures B81B1/00; sequencing of nucleic acids C12Q1/68
• CPC Class Codes: G16B 40/00 ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• CPC Class Codes: G16B 40/20 Supervised data analysis
• CPC Class Codes: G16B 45/00 ICT specially adapted for bioinformatics-related data visualisation, e.g. displaying of maps or networks

Abstract

Techniques for measuring sequences of nucleic acids are provided. Time-based measurements (e.g., forming a histogram) particular to a given sequencing cell can be used to generate a tailored model. The model can include probability functions, each corresponding to different states (e.g., different states of a nanopore). Such probability functions can be fit to a histogram of measurements obtained for that cell. The probability functions can be updated over a sequencing run of the nucleic acid so that drifts in physical properties of the sequencing cell can be compensated. A hidden Markov model can use such probability functions as emission probabilities for determining the most likely nucleotide states over time. For sequencing cells involving a polymerase, a 2-state classification between bound and unbound states of the polymerase can be performed. The bound regions can be further analyzed by a second classifier to distinguish between states corresponding to different bound nucleotides.

Claims

1. A method of using a sequencing cell, the method comprising:

obtaining a first set of signal values measured from a nucleic acid over a first time interval for a sequencing cell, wherein the first set of signal values includes measurements for each of four cell states of the sequencing cell, the four cell states corresponding to different types of nucleotides;

creating a first histogram of the first set of signal values, the first histogram being a data structure storing a plurality of counts, each count corresponding to a number of signal values within a bin, each bin of the first histogram corresponding to different numerical values;

for each cell state of the four cell states;

determining a probability function that assigns emission probabilities of being in the cell state to the different numerical values, the probability function determined using the plurality of counts for the bins of the first histogram;

determining a transmission matrix providing pairwise transition probabilities between four nucleotide states of the nucleic acid, the four nucleotide states corresponding to the different types of nucleotides;

creating a trellis diagram over T time steps, each time step corresponding to one signal value of the first set of signal values, wherein the trellis diagram at a given time step includes the four nucleotide states, each having an emission probability determined using a probability function of a corresponding cell state, and wherein nucleotide states at one time step are connected to nucleotide states at a next time step in accordance with the pairwise transition probabilities;

determining an optimal path through the trellis diagram based on the emission probabilities and the pairwise transition probabilities to identify a nucleotide state at each time step;

determining bases comprising a sequence of the nucleic acid using the nucleotide states at the T time steps; and

providing the sequence of the nucleic acid.

2. The method of claim 1, wherein the sequencing cell further includes a nanopore and has a voltage applied across the sequencing cell, and wherein the four cell states correspond to pore states of the nanopore.

3. The method of claim 2, wherein the voltage includes an alternating signal having a first portion and a second portion relative to a reference voltage, and wherein the first set of signal values is measured during the first portion of the alternating signal.

4. The method of claim 3, wherein the sequencing cell includes a polymerase for sequencing the nucleic acid, wherein the nucleotide states corresponding to binding states of the polymerase, and wherein the first set of signal values further includes measurements for a fifth cell state of the sequencing cell corresponding to no tag molecule attached to a nucleotide being threaded in the nanopore.

5. The method of claim 4, wherein the nucleotide states correspond to system states that include sub-states of the cell states and binding states of the polymerase, wherein the system states include combinations of an unbound state for the polymerase and threaded states for the pore states.

6. The method of claim 1, further comprising:

obtaining other sets of signal values measured from other nucleic acids for other sequencing cells; and

for each of the other sequencing cells;

creating another histogram from another set of signal values,

determining probability functions specific to the sequencing cell using the first histogram, and

using the probability functions specific to the sequencing cell to determine the bases comprising a sequence of the nucleic acid in the sequencing cell.

7. The method of claim 1, wherein determining the probability functions using the first histogram includes fitting each probability function to a peak in the first histogram.

8. The method of claim 7, wherein the probability functions form a mixture model, and wherein the probability functions are fit to the first histogram as part of a collective optimization procedure.

9. The method of claim 1, wherein the optimal path through the trellis diagram is determined using a Viterbi decoder.

10. The method of claim 1, further comprising:

determining second probability functions corresponding to a second time interval, the probability functions and the second probability functions forming a set of time-dependent probability functions, wherein the second probability function is determined using the probability functions and a second histogram determined from a second set of signal values measured from the nucleic acid over the second time interval for the sequencing cell,

wherein the bases comprising the sequence of the nucleic acid are determined using the set of time-dependent probability functions.

11. The method of claim 1, wherein the sequencing cell includes a polymerase for sequencing the nucleic acid, wherein the first set of signal values includes measurements for each of five binding states of the sequencing cell, wherein four binding states correspond to different types of nucleotides and a fifth binding state corresponds to no nucleotide being in an active site of the polymerase, wherein the four binding states collectively correspond to a bound state and the fifth binding state is an unbound state, the method further comprising:

classifying each of the first set of signal values as corresponding to the bound state or to the unbound state using a first classifying procedure, wherein the first classifying procedure is a 2-state classifier; and

identifying a subset of signal values as corresponding to the bound state, wherein the trellis diagram is used to determine the nucleotide states corresponding to the subset of signal values.

12. The method of claim 1, wherein the sequencing cell includes a polymerase attached to a nanopore for sequencing the nucleic acid, and wherein obtaining the first set of signal values includes:

applying a voltage across the sequencing cell, wherein the voltage includes an alternating signal having a first portion and a second portion relative to a reference voltage,

wherein at least a portion of the first set of signal values are measured during the first portion of the alternating signal when a tag molecule is threaded in the nanopore of the sequencing cell, the tag molecule corresponding to a particular nucleotide.

13. A method of using a sequencing cell, the method comprising:

obtaining a first set of signal values measured from a nucleic acid over a first time interval for a sequencing cell, wherein the first set of signal values includes measurements for each of four cell states of the sequencing cell, the four cell states corresponding to different types of nucleotides;

creating a first histogram of the first set of signal values, the first histogram being a data structure storing a plurality of counts, each count corresponding to a number of signal values within a bin, each bin of the first histogram corresponding to different numerical values;

for each cell state of the four cell states;

obtaining an initial probability function that assigns emission probabilities of being in the cell state to the different numerical values; and

using the initial probability function and the first histogram to determine a first probability function that assigns emission probabilities of being in the cell state to the different numerical values, the first probability function corresponding to the first time interval;

determining second probability functions corresponding to a second time interval, the first probability functions and the second probability functions forming a set of time-dependent probability functions, wherein the second probability function is determined using the first probability functions and a second histogram determined from a second set of signal values measured from the nucleic acid over the second time interval for the sequencing cell;

determining bases comprising a sequence of the nucleic acid using the set of time-dependent probability functions; and

providing the sequence of the nucleic acid.

14. The method of claim 13, wherein the initial probability function is determined using signal values measured from one or more other sequencing cells.

15. The method of claim 13, wherein the initial probability function is determined using signal values measured from an earlier time interval than the first time interval.

16. The method of claim 13, wherein the initial probability function is determined using signal values measured over a larger time interval that includes the first time interval and other time intervals in sequencing the nucleic acid.

17. The method of claim 16, further comprising:

determining an initial sequence of initial basecalls of the nucleic acid using the initial probability function, wherein initial basecalls have quality scores; and

selecting the first set of signal values based on the quality scores of corresponding initial basecalls.

18. The method of claim 13, wherein the first probability functions are used as prior distributions for determining the second probability functions as posterior distributions using Bayesian statistics.

19. The method of claim 13, wherein determining the bases using the set of time-dependent probability functions comprises:

determining a transmission matrix providing pairwise transition probabilities between four nucleotide states of the nucleic acid, the four nucleotide states corresponding to different nucleotides of the nucleic acid;

creating a trellis diagram over T time steps, each time step corresponding to one signal value of the first set of signal values, wherein the trellis diagram at a given time step includes the four nucleotide states, each having an emission probability determined using a time-dependent probability function of a corresponding cell state and that corresponds to the given time step, and wherein nucleotide states at one time step are connected to nucleotide states at a next time step in accordance with the pairwise transition probabilities; and

determining an optimal path through the trellis diagram based on the emission probabilities and the pairwise transition probabilities to identify a base call at each time step.