This example is taken from the ProBound web server.

This example produces a single CTCF binding model by training on data from two experiments, SMILE-seq and HT-SELEX, jointly.

import torch

import pyprobound
import pyprobound.plotting

Data specification

alphabet = pyprobound.alphabets.DNA()

url = "http://pbdemo.x3dna.org/files/example_data/multiTF/"
dataframe_SMiLE = pyprobound.get_dataframe(
    f"{url}countTable.0.CTCF_r3.tsv.gz"
)  # SMiLE-seq count table generated from Isakova et al. (2017)
dataframe_SELEX = pyprobound.get_dataframe(
    f"{url}countTable.0.CTCF_ESAJ_TAGCGA20NGCT.tsv.gz"
)  # SELEX count table generated from Jolma et al. (2013)

count_table_SMiLE = pyprobound.CountTable(
    dataframe_SMiLE,
    alphabet,
    left_flank="ACACTCTTTCCCTACACGACGCTCTTCCGATCTTGACGTC",
    right_flank="GACGTCAGATCGGAAGAGCTCGTATGCCGTCTTCTGCTTG",
    left_flank_length=5,
    right_flank_length=5,
)
count_table_SELEX = pyprobound.CountTable(
    dataframe_SELEX,
    alphabet,
    left_flank="CGGAGTCGGCAAGCAGAAGACGGCATACGATAGC",
    right_flank="TCGCTAGATCGGAAGAGCGTCG",
    left_flank_length=5,
    right_flank_length=5,
)
count_tables = [count_table_SMiLE, count_table_SELEX]

Model specification

PSAMs

nonspecific = pyprobound.layers.NonSpecific(alphabet=alphabet, name="NS")
psam_bias = pyprobound.layers.PSAM(
    kernel_size=1, alphabet=alphabet, score_reverse=False
)
psam_1 = pyprobound.layers.PSAM(
    kernel_size=12,
    alphabet=alphabet,
    max_kernel_size=18,
    increment_flank=True,
    shift_footprint_heuristic=True,
    increment_footprint=True,
    increment_flank_with_footprint=True,
)
psam_2 = pyprobound.layers.PSAM(
    kernel_size=12,
    alphabet=alphabet,
    max_kernel_size=18,
    increment_flank=True,
    shift_footprint_heuristic=True,
    increment_footprint=True,
    increment_flank_with_footprint=True,
)

Modes

mode_nonspecific_SMiLE = pyprobound.Mode.from_nonspecific(
    nonspecific, count_table_SMiLE
)
mode_bias_SMiLE = pyprobound.Mode.from_psam(
    psam_bias, count_table_SMiLE, train_posbias=True
)
mode_1_SMiLE = pyprobound.Mode.from_psam(psam_1, count_table_SMiLE)
mode_2_SMiLE = pyprobound.Mode.from_psam(psam_2, count_table_SMiLE)
modes_SMiLE = [
    mode_nonspecific_SMiLE,
    mode_bias_SMiLE,
    mode_1_SMiLE,
    mode_2_SMiLE,
]

mode_nonspecific_SELEX = pyprobound.Mode.from_nonspecific(
    nonspecific, count_table_SELEX
)
mode_bias_SELEX = pyprobound.Mode.from_psam(
    psam_bias, count_table_SELEX, train_posbias=True
)
mode_1_SELEX = pyprobound.Mode.from_psam(psam_1, count_table_SELEX)
mode_2_SELEX = pyprobound.Mode.from_psam(psam_2, count_table_SELEX)
modes_SELEX = [
    mode_nonspecific_SELEX,
    mode_bias_SELEX,
    mode_1_SELEX,
    mode_2_SELEX,
]

Rounds

rounds_SMiLE = pyprobound.rounds.repeat_round(
    modes_SMiLE, 2, pyprobound.rounds.BoundUnsaturatedRound
)
rounds_SELEX = pyprobound.rounds.repeat_round(
    modes_SELEX, 5, pyprobound.rounds.BoundUnsaturatedRound
)

Experiments

experiment_SMiLE = pyprobound.Experiment(
    rounds_SMiLE,
    name="CTCF-SMiLE",
    counts_per_round=count_table_SMiLE.counts_per_round,
)
experiment_SELEX = pyprobound.Experiment(
    rounds_SELEX,
    name="CTCF-SELEX",
    counts_per_round=count_table_SELEX.counts_per_round,
)
experiments = [experiment_SMiLE, experiment_SELEX]

Model

model = pyprobound.MultiExperimentLoss(experiments, pseudocount=20)

Fitting

optimizer = pyprobound.Optimizer(
    model,
    count_tables,
    greedy_threshold=2e-4,
    device="cpu",
    checkpoint="CTCF_multiexp.pt",
    output="CTCF_multiexp.txt",
)

optimizer.train_sequential()

tensor(1.0221)

optimizer.reload()

{'time': 'Wed Apr 24 02:23:43 2024',
 'version': '1.3.1',
 'flank_lengths': ((13, 13), (13, 13))}

Loss

with torch.inference_mode():
    loss, reg = model(count_tables)
    print(loss, reg, loss + reg)

tensor(1.0040) tensor(0.0181) tensor(1.0221)

Position bias

pyprobound.plotting.posbias(mode_bias_SMiLE)
pyprobound.plotting.posbias(mode_bias_SELEX)
../_images/f0d1fa4500562abf15011045e33ec9ac0f69125e7def570076c8796ffe0e2614.png ../_images/ffc1846f9065d8b74d9aeda32fd5d2664902a3d300cd0a80fe6c172dffe4b819.png

Probe enrichment

for experiment, count_table in zip(experiments, count_tables):
    pyprobound.plotting.probe_enrichment(experiment, count_table)
../_images/991f801c47dda8f8777a2cd1b3f2e047dfe5c9a727d35397606e7f104b92e383.png ../_images/1129d8431daa12126d4acbbf6138eb4f5c2c2b354d7c657da0a6aac30eab3b30.png

Mode contribution

for experiment, count_table in zip(experiments, count_tables):
    pyprobound.plotting.contribution(experiment.rounds[-1], count_table)
../_images/f603a7ecc26d55a30f1e41baede9b8e27bdf960b441109ecb833abc6eb3970ff.png ../_images/34c6c31f562978d3ffd596e97f5da47191593d6badd44132d3dffb62deee60fb.png