We have compared Tachyon's performance to DIAMOND and BLAST. As a reference we used NCBI NR (123GB), and have used three query sets (HumDiv, HumVar, Escherichia).
To measure sensitivity we used SW# as reference. The following tables shows sensitivity:
HumDiv:
| RANK | Tachyon | Tachyon sensitive | Diamond | Diamond sensitive | Diamond more sensitive | BLAST |
|---|---|---|---|---|---|---|
| 1 | 100.00% | 100.00% | 70.79% | 70.79% | 70.79% | 76.51% |
| 5 | 100.00% | 100.00% | 86.92% | 87.05% | 87.05% | 90.79% |
| 10 | 100.00% | 100.00% | 89.37% | 89.75% | 89.75% | 94.10% |
| 25 | 99.99% | 100.00% | 91.61% | 92.09% | 92.09% | 96.24% |
| 50 | 99.63% | 99.94% | 92.60% | 93.12% | 93.12% | 96.21% |
| 75 | 98.82% | 99.84% | 93.26% | 93.79% | 93.79% | 96.25% |
| 100 | 98.12% | 99.62% | 94.12% | 94.57% | 94.57% | 96.56% |
| 150 | 96.50% | 99.12% | 95.37% | 95.90% | 95.90% | 96.66% |
| 200 | 94.50% | 98.26% | 96.05% | 96.68% | 96.68% | 96.40% |
| 300 | 88.49% | 94.84% | 95.91% | 96.73% | 96.73% | 95.68% |
| 400 | 81.63% | 90.55% | 95.37% | 96.25% | 96.25% | 94.72% |
| 500 | 75.15% | 86.57% | 95.15% | 96.06% | 96.06% | 94.13% |
HumVar:
| RANK | Tachyon | Tachyon sensitive | Diamond | Diamond sensitive | Diamond more sensitive | BLAST |
|---|---|---|---|---|---|---|
| 1 | 99.74% | 99.94% | 79.56% | 79.56% | 89.22% | 83.38% |
| 5 | 99.69% | 99.95% | 89.12% | 89.22% | 91.11% | 92.74% |
| 10 | 99.59% | 99.91% | 90.99% | 91.11% | 93.33% | 94.68% |
| 25 | 99.07% | 99.62% | 93.14% | 93.33% | 93.02% | 96.32% |
| 50 | 96.99% | 98.50% | 92.76% | 93.02% | 93.22% | 95.80% |
| 75 | 94.70% | 97.43% | 92.91% | 93.21% | 94.16% | 95.74% |
| 100 | 92.78% | 96.50% | 93.83% | 94.15% | 95.70% | 95.94% |
| 150 | 89.24% | 94.70% | 95.27% | 95.70% | 95.90% | 96.02% |
| 200 | 85.17% | 92.19% | 95.33% | 95.89% | 95.68% | 95.47% |
| 300 | 76.52% | 86.15% | 94.74% | 95.66% | 95.20% | 94.43% |
| 400 | 68.67% | 80.44% | 93.89% | 95.18% | 94.51% | 93.52% |
| 500 | 61.81% | 75.09% | 92.77% | 94.47% | 92.58% |
Escherichia:
| RANK | Tachyon | Tachyon sensitive | Diamond | Diamond sensitive | Diamond more sensitive | BLAST |
|---|---|---|---|---|---|---|
| 1 | 99.88% | 99.92% | 40.40% | 40.42% | 40.42% | 52.85% |
| 5 | 99.45% | 99.41% | 51.63% | 51.97% | 51.97% | 58.64% |
| 10 | 98.94% | 98.90% | 58.18% | 58.66% | 58.66% | 64.52% |
| 25 | 98.11% | 98.04% | 67.81% | 68.53% | 68.53% | 73.65% |
| 50 | 97.18% | 97.11% | 75.43% | 76.31% | 76.31% | 81.22% |
| 75 | 96.44% | 96.44% | 79.67% | 80.67% | 80.67% | 85.47% |
| 100 | 95.81% | 95.89% | 82.41% | 83.50% | 83.50% | 88.13% |
| 150 | 94.78% | 95.01% | 85.76% | 86.95% | 86.95% | 91.34% |
| 200 | 93.92% | 94.22% | 87.73% | 89.04% | 89.04% | 93.33% |
| 300 | 92.31% | 92.64% | 89.74% | 91.26% | 91.26% | 95.34% |
| 400 | 90.64% | 90.99% | 90.50% | 92.21% | 92.21% | 96.27% |
| 500 | 89.03% | 89.38% | 90.84% | 92.69% | 92.69% | 96.79% |
To measure sensitivity of BLASTX mode we used 5K reads of Human genome and align it against NR database. For each read we measure how many times this read is mapped to Human reference in first N (1, 5, 10) results. From the results we can see that Tachyon in sensitive mode are better than Diamond for N < 5.
| N | Tachyon | Tachyon Sensitive | Diamond | BLAST |
|---|---|---|---|---|
| 1 | 615 | 769 | 731 | 937 |
| 5 | 1915 | 3076 | 3083 | 3719 |
| 10 | 2945 | 5156 | 5740 | 6649 |
To measure speed we have used same query sets as for sensitivity. Those tables show that tachyon is faster than Diamond, and Tachyon in sensitivity mode is also faster than Diamond in sensitivity mode.
| Tachyon | Tachyon sensitive | Diamond | Diamond sensitive | Diamond more sensitive | |
|---|---|---|---|---|---|
| HumDiv | 197s | 755.347s | 629.359s | 2596.23s | 2578.82s |
| HumVar | 738.115s | 1708.885s | 830.195s | 3530.22s | 3588.32s |
| Esherichia | 435.262s | 1179.694s | 834.69s | 3651.16s | 3617.92s |
To build run the following commands from the cloned project:
mkdir -p build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..ma
makeAfter running those commands, the executable Tachyon will
be into build directory.
If you want to use this library with some Java app run cmake as :
cmake -DCMAKE_BUILD_TYPE=Release -DBUILD_NATIVE=ON ..
before make.
For this example we will use Astral database (https://scop.berkeley.edu/downloads/scopeseq-2.07/astral-scopedom-seqres-gd-sel-gs-bib-40-2.07.fa).
You can find it into db folder.
The first step is to create a indexes database. From the build folder run:
./Tachyon makedb --in ../db/astral.fa --db ./indexes.bin
After that searching can be performed:
./Tachyon blastp --db indexes.bin --query ../db/queries.fa --out output
If you want to use the NR database, first download it:
curl 'https://ftp.ncbi.nlm.nih.gov/blast/db/FASTA/nr.gz' --output nr.gz
Once you get nr.gz file extract it using gzip -d nr.gz command.
After that make indexes:
./Tachyon makedb --in ${PATH_TO_FOLDER_OF_NR_DB}/nr --db ${PATH_TO_FOLDER_OF_NR_DB}/nr.indexes
and search using the same query file as it was for Astral database:
./Tachyon blastp --db ${PATH_TO_FOLDER_OF_NR_DB}/nr.indexes --query ../db/queries.fa --out nr_output
Tachyon also suports mapping RNA reads against protein database.
./Tachyon blastx --db ${PATH_TO_FOLDER_OF_NR_DB}/nr.indexes --query ../db/human_genome.fa --out blastx_output
This work has been developed under supervision of prof. Mile Šikić (collab with Sebastian Maurer-Stroh from Bioinformatics Institute, A*STAR). I would like to acknowledge the contribution of Robert Vaser from the Laboratory for Bioinformatics and Computational Biology for his help on this project. Both of them are also members of AI in Genomics, Genome Institute of Singapore, A*STAR