The Science How gene expression becomes an instrument
Arabidopsis thaliana is a plant that scientists love to study. It has a small genome and a fast reproduction cycle, but is still a complex flowering plant. These qualities makes it an ideal candidate for research.
Arabidopsis contains around 27,000 genes. Many of these genes can modulate their activities based on stimuli. Yuchen Yang and colleagues wanted to study the effect of day/night cycle on gene activities.
They grew Arabidopsis under 16hr light and 8hr dark cycles, and measured activities of each of the 27,000 genes every 3 hours over 2 days: totaling 16 data points per gene.
Here is TOC1 (AT5G61380), a known circadian gene:
We can “hear” the data if we treat the resulting plot as an audio waveform.
An audio waveform is a representation of sound:
For the synthesizer below, amplitude is locked and frequency is mapped to the keyboard, so you can focus on waveform shape. That gives about 27,000 “instruments” derived from the Arabidopsis genome
Use the synthesizer to explore the Arabidopsis transcriptional soundscape.
Assign to a track
Methods
Processing Raw Reads
The Diurnal Dataset from Yuchen Yang and colleagues is downloaded from GSE137732 . These are bulk RNAseq dataset from Arabidopsis grown under long-day conditions [16hr light / 8hr dark]
The paired end reads were processed with fastp with "--detect_adapter_for_pe". Results before and after trimming were checked using Multiqc. Fragments were mapped to TAIR10 transcriptome and transcripts were counted using salmon using the genome as decoy.
DESeq2: adjusting for difference in sequencing depth and RNA composition
DESeq2 was used to normalize the samples after using tximport to summarize transcript level data. Normalized counts is used for sound wave generation: FFT analysis and as input waveform for the synth. Genes with less than 5 average DESeq2 normalized counts across all samples were dropped.TPM: adjusting for difference in gene lengths
Log10 of mean TPM on DESeq2 normalized counts were kept for downstream between gene comparisons.VST: adjusting for heteroskedasticity
VST transformation was performed and the output was used for analysis of the biological signal: UMAP, LombScargle and Gaussian Process modeling.UMAPs
Counts
VST data for each gene is z-score normalized across the 16 timepoints before running UMAP with n_neighbors = 10, min_dist = 0.25, and metric = "cosine". Distance between points is (1 - Pearson correlation).
FFT
Each gene's normalized counts were interpolated using Catmull-Rom, mean centered, and max-absolute scaled to between -1 and 1. FFT is performed on each waveform and the resulting magnitude of each detected frequencies within a gene are converted to dB relative to the strongest frequency and floored at -40dB by setting any frequency 100x weaker than the strongest frequency to 0. All dB values were shifted up by 40dB so the data ranges from 0~40dB. The resulting array is used to generate the UMAP with n_neighbors = 10, min_dist = 0.25, and metric = "euclidean"
Limitations
- Expression strength is currently not mapped to anything sonically, and is only used as a filter.
- 16 points is limiting in terms of possible sounds and harmonics.
- The same smoothing function (Catmull-Rom) used on the waveform makes it so they all sound sonically in the same family.