In vitro selection is the experiment process by which random sequences of DNA or RNA are examined in parallel to identify those particular sequences (i.e., active sequences) that have a desired function. In vitro selection is widely used for the identification of novel binding and catalytic activity. Even though the initial sequence throughput of in vitro selection is normally ~10^14, the number of eventually identified active sequences is usually on the order of 10^1. Such dramatic difference between the throughputs of input and output for selection experiments is largely attributed to the lack of sequence-function relationship. In other words, it is not possible to screen each sequence for its activity, albeit state-of-the-art sequencing technology allows us to discover more than 10^6 sequences per run.

In this repository, an LDA-based algorithmic method, ASPECT, is developed to allow prediction of active sequences without sequence-activity information.

The following repositories are required.

1. pandas:

   pip install pandas

2. gensim:

   pip install gensim

To run ASPECT

python predictor.py [-h] [-t TOKEN_SIZE] [-n NUM_TOPICS] path

positional arguments:

​ path folder path to the sequencing data

optional arguments:

​ -h, --help show this help message and exit

​ -k TOKEN_SIZE, --token_size TOKEN_SIZE token size to parse DNA/RNA squences (default is 8)

​ -rn ROUND_NUMBER, --round_number ROUND_NUMBER round number of selection pool to predict active sequences (default is the latest round of the input)

​ -tn NUM_TOPICS, --num_topics NUM_TOPICS number of topics in the topic model (default is 10)

Input

The path in the above command should be a folder path containing all the sequencing data for a selection experiment. In this folder, each round of sequencing data will be stored in a csv file named as "roundX.csv", where X stands for the round number. To run the active sequence predictor, at least two rounds of sequencing data will be needed.

In each csv file, sequencing data should be stored with column name "sequence".

Output

ASPECT will output all predicted active sequences found in the last round csv file, separated by different cluster number. Each csv file is composed by three columns: seq_num, confidence, and sequence.

seq_num represents the index 1 of a specific sequence in the last round csv file.

confidence represents the confidence of claiming that a specific sequence belongs to certain active sequence cluster.

sequence represents the contain of the active sequence.

In this repository, NGS data for round 4-7 in Diels-Alderase (DAse) ribozyme selection experiment was used as an example in the folder ./DAse.

To predict active DAse sequences from the example data

python predictor.py /path/to/repository/DAse

NGS data for DAse selection was originally reported by

Ameta S, et al. Nucleic Acids Res. 2014, 42, 1303