Compressed sensing for screening and tissue imaging

1. A method of imaging genes in situ comprising:

(a) preparing a probe set for each gene module in a set of gene modules, wherein the probe set comprises a plurality of probes to a sub-set of genes that are representative of all genes in a given gene module;

(b) preparing composite probe sets, wherein each composite probe set comprises a different combination of probes from the plurality of probes prepared in step (a);

(c) obtaining a set of images of a tissue sample, wherein, for each image in the set of images, a different composite probe set is used and imaged simultaneously;

(d) encoding, using a pretrained encoder, an encoded representation of each image in the set of images;

(e) decompressing the encoded representation of each image by solving a sparse optimization problem that estimates gene module activities and individual gene abundances given the composite probe set and the gene modules used; and

(f) generating, using a pretrained decoder, a new set of images from the encoded representation of each image after the decompressing, the new set of images providing a spatial expression pattern for each gene represented in the composite probe set.

2. The method of claim 1 , further comprising:

identifying the set of gene modules by collecting training data from a tissue or cell by performing single cell sequencing.

3. The method of claim 2 , wherein the single cell sequencing comprises whole transcriptome amplification.

4. The method of claim 2 , wherein the training data is obtained across a library of cells.

5. The method of claim 2 , wherein the cell comprises a eukaryotic cell.

6. The method of claim 5 , wherein the eukaryotic cell is a mammalian cell.

7. The method of claim 4 , wherein the library of cells comprises multiple cell types.

8. The method of claim 1 , wherein the tissue sample comprises or is from a biopsy from a subject.

9. The method of claim 1 , further comprising:

identifying the set of gene modules based upon a random sampling of one or more composite genes.

10. The method of claim 2 , wherein the tissue or cell is from the motor cortex.

11. The method of claim 10 , wherein the motor cortex is a mouse motor cortex.

12. The method of claim 1 , wherein the decompressing is by compressed sensing.

13. The method of claim 5 , wherein the eukaryotic cell is a human cell.

14. The method of claim 1 , wherein the tissue sample comprises neurons.

15. The method of claim 1 , wherein the tissue sample is olfactory epithelium.

16. A computer-implemented method to scan in situ images and infer gene abundances, comprising:

(a) receiving a set of images of a tissue sample obtained by simultaneously measuring expression of a composite probe set, wherein for each image in the set of images a different composite probe set is used, and wherein each composite probe set comprises a different combination of gene probes representative of different gene modules;

(b) encoding, using a pretrained encoder, an encoded representation of each image in the set of images;

(c) decompressing the encoded representation of each image by solving a sparse optimization problem that estimates gene module activities and individual gene abundances given the composite probe set and the gene modules used; and

(d) generating a composite set of images that provides a spatial expression pattern for each gene represented in the composite probe set.

17. The method of claim 16 , wherein the method further comprises pre-processing the set of images.

18. The method of claim 17 , wherein pre-processing comprise of normalizing, stitching, aligning, and segmenting the images in each color, field of view, round, and tissue or cells.

19. The method of claim 17 , wherein pre-processing comprises of smoothing and rescaling the one or more in situ image.