Shiny Python: Reactive Isolate doesn't seem to work with my implementation

Hello, reactive.isolate() doesn't seems to be working for me. I'm wondering if it's a bug in shiny or if there's something wrong with my code.

In my code, I'm taking two user-defined inputs: a list of genes, and a list of abnormal phenotypes. Ideally, the code wouldn't compute the result until the user presses "compute."

I tried the canonical:

def output_function():
    with reactive.isolate():
         # compute and return result

But it still tries to compute! Can anyone help me figure out what's wrong? Here is my code:

from shiny import App, render, ui, reactive
from pyrsistent import pset, pvector # immutable objects
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
import pandas as pd
import networkx as nx


# Logscale version of the geometric mean
def gmean(data, axis=0):
    return np.exp(np.mean(np.log(data), axis=axis))

# get geometric mean of ranks for given terms
def combine_ranks(data, hpo_terms):

    # Make sure hpo_terms are in the data
    hpo_terms = [x for x in hpo_terms if x in data.index]

    # Take the vertical mean (across scores in each gene)
    # Label as genes
    return pd.Series(gmean(data.loc[hpo_terms, :], axis=0), index=data.columns)

# given a list of hpo terms, get the closest ancestor in dataset columns
def reduce_terms(data, hpo_tree, hpo_terms):
    term_list = set() # sets cannot contain duplicate values (sets are mutable)
    for term in hpo_terms:
        # get the list of terms that actually are in dataset
        # HP:0000118 is "phenotypic abnormality"
        path = [x for x in nx.shortest_path(hpo_tree, term, 'HP:0000118') if x in data.columns]

        if len(path) > 0: # if we've moved towards HP:0000118, add to term list
            print("Phenotype not in model, finding closest ancestor.")
    # Return list, not set
    return list(term_list)

# Loading data

DATADIR = "app_discan/" # Empty for

# Gene and disease annotations
gene_annotation = nx.read_graphml(DATADIR + "gene_annotation.graphml")
disease_net = nx.read_graphml(DATADIR + "disease_annot.graphml")

# HPO Terms
hpo_net = nx.read_graphml(DATADIR + "hp.220616.obo.graphml") 
hpo_ids = [x for x in hpo_net.nodes() if nx.has_path(hpo_net, x, 'HP:0000118')]
hpo_ids = hpo_ids[1:]
hpo_names = [hpo_net.nodes()[x]['name'] for x in hpo_ids]
hpo_terms = ["%s: %s" % (hpo_ids[x], hpo_names[x]) for x in range(len(hpo_ids))]

# Diseases
disease_ids = [x for x in disease_net.nodes() if disease_net.nodes()[x]['type'] == 'disease' and 'name' in disease_net.nodes()[x]]
disease_names = [disease_net.nodes()[x]['name'] for x in disease_ids if 'name' in disease_net.nodes()[x]]
diseases = ["%s: %s" % (disease_ids[x], disease_names[x]) for x in range(len(disease_ids))]

# Model

# Read from chunks
model_probas = pd.read_csv(DATADIR + "compr_probas_2022_00.csv.gz", index_col=0, compression='gzip')
for chunk in range(1, 10):
    filename = DATADIR + "compr_probas_2022_%02d.csv.gz" % chunk
    model_probas = pd.concat(
        [model_probas, pd.read_csv(filename, index_col=0, compression='gzip')],

# Create rank matrix
model_ranks = model_probas.rank(axis=1, method="min", pct=True, ascending=False)
model_ranks.index = [str(x) for x in model_ranks.index]

# UI
app_ui = ui.page_fluid(

            ui.markdown("##### Welcome"),
                "This application calculates the likelihood of involvement of one or " \
                "more genes in a genetic syndrome characterized by a set of abnormal phenotypes."

# Genes
            ui.markdown("##### Enter a list of genes: "),
# Diseases
            ui.markdown("##### Select abnormal phenotypes:"),
                "Select a disease, then press add:",
                selected=["OMIM:147920: KABUKI SYNDROME 1: KABUK1"], # Doesn't work for some reason
# Phenotypes
                "OR, Manually select abnormal phenotypes:",

            # ui.output_table("genes_report_table"),
            ui.input_switch("quantile_switch","Plot quantiles"),
            ui.input_action_button("compute", "Compute"),
            ui.markdown("#### Results"),
            # ui.HTML("<hr><b>Warning!</b> The following items were excluded from the analysis:<br>"),

def server(input, output, session):

# Genes
    def get_input_genes():

        # 1. Parse input
        genes = input.genes_list().split() # input genes
        genes = [x.upper() for x in genes] # cast to upper
        genes = list(set(genes)) # unique elements only

        # 2. Check if genes are in annotation
        genes_out = [g for g in genes if not g in gene_annotation.nodes()]
        if len(genes_out) != 0:
            # remove from genes to consider
            genes = [g for g in genes if not g in genes_out]

        df = pd.DataFrame(genes, columns=["Input"])

        # 3. Get Entrez IDs of considered genes

        # A long lambda function:
        # - if the input is already an Entrez ID, keep that.
        # - else, decode the graph at the given node to find the entrez ID

        df["Entrez ID"] = df["Input"].apply(lambda g: g if gene_annotation.nodes()[g]['type'] == 'entrez' else list(gene_annotation.neighbors(g))[0])

        # 4. Get gene symbols
        df["Gene Symbol"] = df["Entrez ID"].apply(lambda x: list(gene_annotation.neighbors(x))[0])

        # 5. Check that gene is in model
        df["In Model"] = df["Entrez ID"].apply(lambda x: "Y" if x in model_ranks.columns else "N")

        # 7. Reorder Columns
        df = df[['Input', 'Entrez ID', "Gene Symbol", 'In Model']]
        return df, genes_out

    # May choose to display the input genes dataframe on the page
    def genes_report_table():
        df, _= get_input_genes()
        return df

    def warning():

        if input.compute() == 0:
            return ""

        df, invalid_input = get_input_genes()
        not_in_model = df[df["In Model"] == "N"]

        # Report excluded genes
        if len(invalid_input) == 0 and len(not_in_model) == 0:
            return ""

        spool = "Warning! The following genes were excluded from analysis:\n\n"
        if len(invalid_input) != 0:
            for gene in invalid_input:
                spool += f" - {gene} is not an annotated gene. \n"

        if len(not_in_model) != 0:
            for gene in not_in_model.iterrows():
                spool += f" - {gene[1][2]} (Entrez ID {gene[1][1]}) was not included in the model. \n"

        ui.notification_show("Warning! Some genes were excluded from analysis!", type="Warning")
        return spool

# Display HPO selection
    def decode_OMIM():

        # Get OMIM ID
        diseases = [d[:11] for d in input.disease_selected()]

        hp_sel = pvector()
        for dis in diseases:
            # Get hp terms
            hp_map = [x for x in disease_net.neighbors(dis) if nx.has_path(hpo_net, x, 'HP:0000118')]
            hp_names = [hpo_net.nodes()[x]['name'] for x in hp_map]

            # Append ID: PHENOTYPE to selection
            hp_map_name = pvector("%s: %s" % (hp_map[x], hp_names[x]) for x in range(len(hp_map)))
            hp_sel += hp_map_name

        return hp_sel

    def get_hpo_terms():
        # Update hpo list according to selected disease
        if input.add() != 0:
            return set(decode_OMIM() + input.hpo_selected())
            return input.hpo_selected()

    # On Add
    def _():
        if input.add() != 0:
            with reactive.isolate(): # inside this block, don't depend on inputs!
                    label="OR: Manually select abnormal phenotypes:",

    # On clear
    def _():
        if input.clear() != 0:
            with reactive.isolate():
                    label="OR: Manually select abnormal phenotypes:",

# Rank calculations
    def rank_all_genes():
        # Compute the ranks for all genes in model
        with reactive.isolate():

            if input.compute() == 0:
                return None

            # Get input hpo_terms, slice to only HP ID
            terms = get_hpo_terms()

            if len(terms) == 0:
                ui.notification_show("Error: please enter at least one phenotype", type="error")
                return pd.Series(None)

            terms = [x[:10] for x in terms]

            # Selected terms that are in the dataset, or nearest ancestors
            subset_terms = reduce_terms(model_ranks, hpo_net, terms)

            # Perform geometric mean operation on all genes in model
            return combine_ranks(model_ranks, subset_terms)

    def calc_result():
        # Select ranks of chosen genes, return dataframe
        with reactive.isolate():

            if input.compute() == 0:
                return None

            # Select input genes
            df, _ = get_input_genes()
            genes = list(df[df["In Model"]=="Y"]["Entrez ID"])

            # Perform geometric mean operation on all genes in model
            all_ranks = rank_all_genes()

            if len(all_ranks) == 0:
                return None

            # Calculate quantiles
            quantiles = pd.qcut(all_ranks, 5, labels=False)

            # Selected gene ranks, sorted by score
            sel_gene_ranks = all_ranks[genes]
            sel_gene_ranks.sort_values(ascending=True, inplace=True)
            sel_quantiles = quantiles[genes]
            sel_quantiles.sort_values(ascending=True, inplace=True)

            ids = []
            names = []
            symbols = []
            for gene in sel_gene_ranks.index:

            columns = ["Entrez ID", "Gene Symbol", "Name and Description", "Score", "Quantile"]
            return pd.DataFrame(
                zip(ids, symbols, names, sel_gene_ranks, sel_quantiles+1),

    def display_result():
        # Display the final dataframe
        with reactive.isolate():

            if input.compute() == 0:
                return None

            return calc_result()

# Plot
    def make_plot():

        # Take a reactive dependency on the compute button

        # But not on any of the other inputs, ideally!
        with reactive.isolate():

            if input.compute() == 0:
                return None

            # Get results
            result = calc_result()
            all_ranks = rank_all_genes()

            if len(all_ranks) == 0:
                return None

            fig, ax = plt.subplots(figsize = (12, 7))

            # Histogram of all scores

            sns.histplot(rank_all_genes(), ax=ax, edgecolor='white')

            # Draw quantile lines
            if input.quantile_switch():
                q = ['25%','50%', '75%']
                colors = ['black', 'black', 'black']
                desc = all_ranks.describe()
                for i in range(len(q)):
                    ax.axvline(desc[q[i]], color=colors[i], ls="--")

            # Draw stems with labels
            stem_y =  np.linspace(600, 1800, result.shape[0])
                linefmt="C3:", markerfmt="C3.", basefmt="None"

            for i in range(result.shape[0]):
                    result["Gene Symbol"][i],
                    xy=(result["Score"][i], stem_y[i]+20),


            plt.title("Distribution of Gene Ranks of all Genes in Model", y=1.02, weight=1.2)

            return ax

    def plot():
        with reactive.isolate():

            if input.compute() == 0:
                return None

            ax = make_plot()
            return ax

app = App(app_ui, server)

1 Like

Sorry for the delay in responding!

reactive.isolate() means "ignore any reactive reads inside here when determining when to re-execute". So your short canonical snippet effectively means, "only recalculate when the button is pushed". What it does not do is, "...and don't calculate until the button is pressed the first time".

The way you can do that is by changing the line input.button() to req(input.button()). This is effectively saying "if button has not been pressed before, silently throw an exception".

However, this pattern of "req(button) plus reactive.isolate() of all the remaining logic" is such a common idiom, that we created a decorator to make it easier: @reactive.event. You can see an example here: Shiny examples

It looks like you have a cool app here, but I'm not able to run it without your data. Any chance that data is publicly available? I'd love to see it in action!

1 Like

Hi jcheng,

Thanks for the concise refresher on @reactive.event. I had gone through the getting started tutorial on shiny for python, so I was aware of that decorator. But I could never get it to work right in my app! Adding that to my code, I noticed that I was missing the @ for some of my @reactive.Calc's.

 reactive.Calc -> @reactive.Calc

Your solution plus my correction seem to have made all the difference. Now, my app works as intended.

If you're interested, the app and the accompanying data are hosted at GitHub - edoardodraetta/discan. It is the implementation of a concept central to our recently submitted paper.

Thanks for your kind and helpful response.

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.