Molecular diagnostic methods for classifying biological samples based on cell density

Abstract

This method enables the estimation of tumors in biological samples by analyzing the gene expression of a predefined list of genes specific to cellular tumor states using a trained binary classifier algorithm.

Claims

1. A computer-based method for classifying biological samples, - The step of receiving a biopsy sample from a subject suspected of having cancer, - A step of receiving instructions for the tumor to be examined in the aforementioned biopsy sample, - A step of extracting RNA and generating a library of polyA+ sequence fragments from the RNA of the sample, - A step of obtaining an output file of the biopsy sample, which includes a sequencer, followed by alignment to a reference genome and counting of fragments or reads, and which includes multiple counts of the appearance of the corresponding gene represented by the polyA+ fragment in the sample, wherein each count defines the expression level of the corresponding gene, - The classification as non-tumor or tumorous cancer is known to be verified by histological cell density testing, and the step of applying a binary classification algorithm trained using a biological training sample with known gene expression levels to the output file, - A method wherein the trained binary classification algorithm is configured to generate a binary estimate of whether the biopsy sample is a tumor analogue or a nontumor of the tumor to be examined, based on the expression levels in the output file of the biopsy sample of a predefined list (List 1) of genes whose expression levels indicate the cell density of the tumor to be examined.

2. The following training steps for the first binary classification algorithm: - A step of extracting RNA from each of the aforementioned in vivo training samples of tumor and non-tumor tissue, and generating a library of polyA+ sequence fragments from the RNA for each of a plurality of first training samples classified as non-tumor (NAT) and tumor (TUM), wherein each first tumor training sample preferably has a level of cell density known by histological examination, and each non-tumor training sample (NAT) is non-tumor tissue adjacent to the tissue of the corresponding tumor training sample (TUM), - A step of processing the sample through a sequencer, followed by alignment to a reference genome and read counting, to obtain a first training output file for each first training sample, which includes multiple counts of occurrences of the corresponding gene represented by the polyA+ fragment in the sample, where each count defines the expression level of the corresponding gene, and - For each gene in the first training output file of each training tumor sample, the step of calculating a correlation index value, preferably a Bayesian factor of the expression level having the known cell density of the corresponding training tumor sample, - The step of including in the list a gene based on the corresponding correlation index value having the known cell density of the corresponding training sample, - A step of training the classification algorithm such that for each training sample (NAT, TUM), the expression levels of each of the genes in the list (List 1) are received as input, and the classification of the training sample as tumor or non-tumor, as known by histological examination, is received as output, wherein the list of genes (List 1) is a list such that the accuracy of the estimation of the classification algorithm applied to the training sample is 95% or higher. The method according to claim 1, including the method described in claim 1.

3. The known cell density level of the biological training sample is at least 2, and the corresponding non-zero known cell density level is associated with the corresponding training output file, which includes the gene expression level. The method according to claim 2, further comprising the step of identifying a first and second plurality of genes based on the correlation index for each of the training samples and each non-zero cell density level, wherein the list (List 1) includes genes from the first and second plurality of genes whose expression levels change with variation in cell density.

4. - A step of applying an unsupervised learning algorithm that takes as input the expression level of genes whose correlation index value is higher than the threshold, and dividing the corresponding training sample into at least two groups, The method according to any one of claims 2 or 3, further comprising the step of increasing the threshold of the correlation index value to match the state of the tumor cells when the separation accuracy between the NAT sample and the TUM sample in each of the groups is lower than a predefined threshold, for example 100%.

5. The method according to any one of claims 1 to 4, wherein the first classifier is a generalized linear model (GLM).

6. The method according to any one of claims 2 to 5, wherein the Bayesian factor is 8 or more, preferably 10 or more.

7. The method according to any one of claims 1 to 6, wherein the library preparation and analysis steps are performed using 3'DGE technology.

8. The method according to any one of claims 1 to 7, wherein the sample is obtained from a breast tissue biopsy.

9. The method according to any one of claims 1 to 8, wherein the sample is a formalin-fixed paraffin-embedded (FFPE) biological tissue sample or an untreated biological tissue sample.

Images