示例#1
0
    def setup_control_area(self):
        # Control area
        self.input_info = widgetLabel(
            widgetBox(self.controlArea, "Info", addSpace=True),
            'No data on input.\n')
        self.custom_gs_col_box = box = vBox(self.controlArea,
                                            'Custom Gene Set Term Column')
        box.hide()

        self.reference_radio_box = radioButtonsInBox(
            self.controlArea,
            self,
            "use_reference_data",
            ["Entire genome", "Reference gene set (input)"],
            tooltips=[
                "Use entire genome (for gene set enrichment)",
                "Use reference set of genes"
            ],
            box="Reference",
            callback=self.invalidate,
        )

        self.reference_radio_box.setEnabled(False)

        gene_sets_box = widgetBox(self.controlArea, "Gene Sets")
        self.gs_widget = GeneSetsSelection(gene_sets_box, self,
                                           'stored_gene_sets_selection')
        self.gs_widget.hierarchy_tree_widget.itemClicked.connect(
            self.update_tree_view)

        self.commit_button = auto_commit(self.controlArea,
                                         self,
                                         "auto_commit",
                                         "&Commit",
                                         box=False)
    def setup_control_area(self):
        # Control area
        self.custom_gs_col_box = box = vBox(self.controlArea, 'Custom Gene Set Term Column')
        box.hide()

        gene_sets_box = widgetBox(self.controlArea, "Gene Sets")
        self.gs_widget = GeneSetsSelection(gene_sets_box, self, 'stored_gene_sets_selection')
        self.gs_widget.hierarchy_tree_widget.itemClicked.connect(self.update_tree_view)

        self.commit_button = auto_commit(self.controlArea, self, "auto_commit", "&Commit", box=False)
class OWClusterAnalysis(OWWidget):
    name = "Cluster Analysis"
    description = "The widget displays differentially expressed genes that characterize the cluster, " \
                  "and corresponding gene terms that describe differentially expressed genes"
    icon = "../widgets/icons/OWClusterAnalysis.svg"
    priority = 100

    class Inputs:
        data_table = Input('Data', Table)
        custom_sets = Input('Custom Gene Sets', Table)

    class Outputs:
        selected_data = Output('Selected Data', Table)
        gene_scores = Output('Gene Scores', Table)
        gene_set_scores = Output('Gene Set Scores', Table)

    class Information(OWWidget.Information):
        pass

    class Warning(OWWidget.Warning):
        gene_enrichment = Msg('{}, {}.')
        no_selected_gene_sets = Msg(
            'No gene set selected, select them from Gene Sets box.')

    class Error(OWWidget.Error):
        no_cluster_indicator = Msg('No cluster indicator in the input data')
        gene_as_attributes = Msg(
            'Genes, in the input data, are expected as column names')
        organism_mismatch = Msg(
            'Organism in input data and custom gene sets does not match')
        cluster_batch_conflict = Msg(
            'Cluster and batch must not be the same variable')

    settingsHandler = ClusterAnalysisContextHandler()
    cluster_indicators = ContextSetting([])
    batch_indicator = ContextSetting(None)
    stored_gene_sets_selection = ContextSetting(tuple())

    scoring_method_selection = ContextSetting(0)
    scoring_method_design = ContextSetting(0)
    scoring_test_type = ContextSetting(0)

    # genes filter
    max_gene_count = Setting(20)
    use_gene_count_filter = Setting(True)

    max_gene_p_value = Setting(0.1)
    use_gene_pval_filter = Setting(False)

    max_gene_fdr = Setting(0.1)
    use_gene_fdr_filter = Setting(True)

    # gene sets filter
    min_gs_count = Setting(5)
    use_gs_count_filter = Setting(True)

    max_gs_p_value = Setting(0.1)
    use_gs_pval_filter = Setting(False)

    max_gs_fdr = Setting(0.1)
    use_gs_max_fdr = Setting(True)

    # auto commit results
    auto_commit = settings.Setting(False)

    custom_gene_set_indicator = settings.Setting(None)

    def __init__(self):
        super().__init__()

        # widget attributes
        self.input_data = None
        self.store_input_domain = None
        self.input_genes_names = []
        self.input_genes_ids = []

        self.tax_id = None
        self.use_attr_names = None
        self.gene_id_attribute = None

        # custom gene set input
        self.feature_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, StringVariable))
        self.custom_data = None
        self.custom_tax_id = None
        self.custom_use_attr_names = None
        self.custom_gene_id_attribute = None
        self.custom_gene_id_column = None
        self.num_of_custom_sets = None

        self.rows_by_cluster = None
        self.rows_by_batch = None
        self.clusters = []
        self.new_cluster_profile = []

        # data model
        self.cluster_info_model = None

        # Info
        info_box = vBox(self.controlArea, 'Info')
        self.input_info = widgetLabel(info_box)

        # Cluster selection
        self.cluster_indicator_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, ), separators=False)
        self.cluster_indicator_box = widgetBox(self.controlArea,
                                               'Cluster Indicator')

        self.cluster_indicator_view = listView(
            self.cluster_indicator_box,
            self,
            'cluster_indicators',
            model=self.cluster_indicator_model,
            selectionMode=QListWidget.MultiSelection,
            callback=self.invalidate,
            sizeHint=QSize(256, 70))

        # Batch selection
        self.batch_indicator_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, ), separators=False, placeholder="")
        box = widgetBox(self.controlArea, 'Batch Indicator')
        self.batch_indicator_combobox = comboBox(
            box,
            self,
            'batch_indicator',
            model=self.batch_indicator_model,
            sendSelectedValue=True,
            callback=self.batch_indicator_changed)

        # Gene scoring
        box = widgetBox(self.controlArea, 'Gene Scoring')
        self.gene_scoring = GeneScoringWidget(box, self)
        self.gene_scoring.set_method_selection_area('scoring_method_selection')
        self.gene_scoring.set_method_design_area('scoring_method_design')
        self.gene_scoring.set_test_type('scoring_test_type')

        # Gene Sets widget
        gene_sets_box = widgetBox(self.controlArea, "Gene Sets")
        self.gs_widget = GeneSetsSelection(gene_sets_box, self,
                                           'stored_gene_sets_selection')
        self.gs_widget.hierarchy_tree_widget.itemClicked.connect(
            self.__gene_sets_enrichment)

        # custom gene sets area
        box = vBox(self.controlArea, "Custom Gene Sets")

        if self.custom_gene_set_indicator not in self.feature_model:
            self.custom_gene_set_indicator = None

        self.gs_label_combobox = comboBox(
            box,
            self,
            "custom_gene_set_indicator",
            sendSelectedValue=True,
            model=self.feature_model,
            callback=self.handle_custom_gene_sets)
        self.gs_label_combobox.setDisabled(True)

        # main area
        splitter = QSplitter(Qt.Horizontal, self.mainArea)
        self.mainArea.layout().addWidget(splitter)

        genes_filter = widgetBox(splitter,
                                 'Filter Genes',
                                 orientation=QHBoxLayout())
        spin(genes_filter,
             self,
             'max_gene_count',
             0,
             10000,
             label='Count',
             tooltip='Minimum genes count',
             checked='use_gene_count_filter',
             callback=self.filter_genes,
             callbackOnReturn=True,
             checkCallback=self.filter_genes)

        doubleSpin(genes_filter,
                   self,
                   'max_gene_p_value',
                   0.0,
                   1.0,
                   0.0001,
                   label='p-value',
                   tooltip='Maximum p-value of the enrichment score',
                   checked='use_gene_pval_filter',
                   callback=self.filter_genes,
                   callbackOnReturn=True,
                   checkCallback=self.filter_genes)

        doubleSpin(genes_filter,
                   self,
                   'max_gene_fdr',
                   0.0,
                   1.0,
                   0.0001,
                   label='FDR',
                   tooltip='Maximum false discovery rate',
                   checked='use_gene_fdr_filter',
                   callback=self.filter_genes,
                   callbackOnReturn=True,
                   checkCallback=self.filter_genes)

        gene_sets_filter = widgetBox(splitter,
                                     'Filter Gene Sets',
                                     orientation=QHBoxLayout())
        spin(gene_sets_filter,
             self,
             'min_gs_count',
             0,
             DISPLAY_GENE_SETS_COUNT,
             label='Count',
             tooltip='Minimum genes count',
             checked='use_gs_count_filter',
             callback=self.filter_gene_sets,
             callbackOnReturn=True,
             checkCallback=self.filter_gene_sets)

        doubleSpin(gene_sets_filter,
                   self,
                   'max_gs_p_value',
                   0.0,
                   1.0,
                   0.0001,
                   label='p-value',
                   tooltip='Maximum p-value of the enrichment score',
                   checked='use_gs_pval_filter',
                   callback=self.filter_gene_sets,
                   callbackOnReturn=True,
                   checkCallback=self.filter_gene_sets)

        doubleSpin(gene_sets_filter,
                   self,
                   'max_gs_fdr',
                   0.0,
                   1.0,
                   0.0001,
                   label='FDR',
                   tooltip='Maximum false discovery rate',
                   checked='use_gs_max_fdr',
                   callback=self.filter_gene_sets,
                   callbackOnReturn=True,
                   checkCallback=self.filter_gene_sets)

        self.cluster_info_view = QTableView()
        self.cluster_info_view.verticalHeader().setVisible(False)
        self.cluster_info_view.setItemDelegate(HTMLDelegate())
        self.cluster_info_view.horizontalHeader().hide()
        self.cluster_info_view.horizontalHeader().setSectionResizeMode(
            QHeaderView.Stretch)

        auto_commit(self.controlArea,
                    self,
                    "auto_commit",
                    "&Commit",
                    box=False)

        self.mainArea.layout().addWidget(self.cluster_info_view)

    def sizeHint(self):
        return QSize(800, 600)

    def __update_info_box(self):
        info_string = ''
        if self.input_genes_ids:
            info_string += '{} samples, {} clusters\n'.format(
                self.input_data.X.shape[0],
                len(self.clusters) if self.clusters else '?')
            info_string += '{:,d} unique genes\n'.format(
                len(self.input_genes_ids))
        else:
            info_string += 'No genes on input.\n'

        if self.custom_data:
            info_string += '{} marker genes in {} sets\n'.format(
                self.custom_data.X.shape[0], self.num_of_custom_sets)

        self.input_info.setText(info_string)

    def __set_cluster_info_model(self):
        self.cluster_info_view.setModel(None)

        self.cluster_info_model = ClusterModel(self)
        self.cluster_info_model.add_rows(self.clusters)

        # add model to the view
        self.cluster_info_view.setModel(self.cluster_info_model)
        # call sizeHint function
        self.cluster_info_view.resizeRowsToContents()
        self.cluster_info_view.selectionModel().selectionChanged.connect(
            self.commit)

    def __create_temp_class_var(self):
        """ See no evil !"""
        cluster_indicator_name = 'Cluster indicators'

        var_index_lookup = dict([(val, idx) for var in self.cluster_indicators
                                 for idx, val in enumerate(var.values)])

        row_profile = None
        new_cluster_values = []

        cart_prod = itertools.product(
            *[cluster.values for cluster in self.cluster_indicators])
        for comb in cart_prod:
            new_cluster_values.append(', '.join([val for val in comb]))
            self.new_cluster_profile.append(
                [var_index_lookup[val] for val in comb])

        row_profile_lookup = dict([(tuple(profile), indx) for indx, (
            profile,
            _) in enumerate(zip(self.new_cluster_profile, new_cluster_values))
                                   ])
        for var in self.cluster_indicators:
            if row_profile is None:
                row_profile = np.asarray(
                    self.input_data.get_column_view(var)[0], dtype=int)
            else:
                row_profile = np.vstack(
                    (row_profile,
                     np.asarray(self.input_data.get_column_view(var)[0],
                                dtype=int)))

        ca_ind = DiscreteVariable.make(
            cluster_indicator_name,
            values=[val for val in new_cluster_values],
            ordered=True)

        domain = Domain(self.input_data.domain.attributes,
                        self.input_data.domain.class_vars,
                        self.input_data.domain.metas + (ca_ind, ))

        table = self.input_data.transform(domain)
        table[:, ca_ind] = np.array(
            [[row_profile_lookup[tuple(row_profile[:, i])]]
             for i in range(row_profile.shape[1])])
        self.input_data = table
        return ca_ind

    def __set_clusters(self):
        self.clusters = []
        self.new_cluster_profile = []
        self.cluster_var = None

        if self.cluster_indicators and self.input_data:

            if isinstance(self.cluster_indicators,
                          list) and len(self.cluster_indicators) > 1:
                self.cluster_var = self.__create_temp_class_var()
            else:
                self.cluster_var = self.cluster_indicators[0]

            self.rows_by_cluster = np.asarray(self.input_data.get_column_view(
                self.cluster_var)[0],
                                              dtype=int)
            for index, name in enumerate(self.cluster_var.values):
                cluster = Cluster(name, index)
                self.clusters.append(cluster)
                cluster.set_genes(self.input_genes_names, self.input_genes_ids)

    def __set_batch(self):
        self.Error.cluster_batch_conflict.clear()
        self.rows_by_batch = None

        if self.batch_indicator == self.cluster_var:
            self.Error.cluster_batch_conflict()
            return
        if self.batch_indicator and self.input_data:
            self.rows_by_batch = np.asarray(self.input_data.get_column_view(
                self.batch_indicator)[0],
                                            dtype=int)

    def __set_genes(self):
        self.input_genes_names = []
        self.input_genes_ids = []

        if self.use_attr_names:
            for variable in self.input_data.domain.attributes:
                self.input_genes_names.append(str(variable.name))
                self.input_genes_ids.append(
                    str(variable.attributes.get(self.gene_id_attribute,
                                                np.nan)))

    def filter_genes(self):
        if self.cluster_info_model:
            # filter genes
            # note: after gene filter is applied, we need to recalculate gene set enrichment
            self.cluster_info_model.apply_gene_filters(
                self.max_gene_p_value if self.use_gene_pval_filter else None,
                self.max_gene_fdr if self.use_gene_fdr_filter else None,
                self.max_gene_count if self.use_gene_count_filter else None)

            # recalculate gene set enrichment
            self.__gene_sets_enrichment()
            # call sizeHint function
            self.cluster_info_view.resizeRowsToContents()

            # commit changes after filter
            self.commit()

    def filter_gene_sets(self):
        if self.cluster_info_model:
            # filter gene sets
            self.cluster_info_model.apply_gene_sets_filters(
                self.max_gs_p_value if self.use_gs_pval_filter else None,
                self.max_gs_fdr if self.use_gs_max_fdr else None,
                self.min_gs_count if self.use_gs_count_filter else None)

            # call sizeHint function
            self.cluster_info_view.resizeRowsToContents()

    def __gene_enrichment(self):
        design = bool(self.gene_scoring.get_selected_desig()
                      )  # if true cluster vs. cluster else cluster vs rest
        test_type = self.gene_scoring.get_selected_test_type()
        method = self.gene_scoring.get_selected_method()
        try:
            if method.score_function == score_hypergeometric_test:
                values = set(np.unique(self.input_data.X))
                if (0 not in values) or (len(values) != 2):
                    raise ValueError('Binary data expected (use Preprocess)')

            self.cluster_info_model.score_genes(
                design=design,
                table_x=self.input_data.X,
                rows_by_cluster=self.rows_by_cluster,
                rows_by_batch=self.rows_by_batch,
                method=method,
                alternative=test_type)
        except ValueError as e:
            self.Warning.gene_enrichment(str(e), 'p-values are set to 1')

    def __gene_sets_enrichment(self):
        if self.input_data:
            self.Warning.no_selected_gene_sets.clear()
            all_sets = self.gs_widget.get_hierarchies()
            selected_sets = self.gs_widget.get_hierarchies(only_selected=True)

            if len(selected_sets) == 0 and len(all_sets) > 0:
                self.Warning.no_selected_gene_sets()

            # save setting on selected hierarchies
            self.stored_gene_sets_selection = tuple(selected_sets)
            ref_genes = set(self.input_genes_ids)

            try:
                self.cluster_info_model.gene_sets_enrichment(
                    self.gs_widget.gs_object, selected_sets, ref_genes)
            except Exception as e:
                # TODO: possible exceptions?

                raise e

            self.filter_gene_sets()

    def invalidate(self, cluster_init=True):
        if self.input_data is not None and self.tax_id is not None:
            self.Warning.gene_enrichment.clear()

            if self.cluster_info_model is not None:
                self.cluster_info_model.cancel()

            self.__set_genes()
            if cluster_init:
                self.__set_clusters()
            self.__set_batch()
            self.__set_cluster_info_model()

            # note: when calling self.__gene_enrichment we calculate gse automatically.
            #       No need to call self.__gene_sets_enrichment here
            self.__gene_enrichment()
            self.__update_info_box()

    def batch_indicator_changed(self):
        self.invalidate(cluster_init=False)

    @Inputs.data_table
    def handle_input(self, data):
        self.closeContext()
        self.Warning.clear()
        self.Error.clear()

        self.input_data = None
        self.store_input_domain = None
        self.stored_gene_sets_selection = tuple()
        self.input_genes_names = []
        self.input_genes_ids = []
        self.tax_id = None
        self.use_attr_names = None
        self.gene_id_attribute = None
        self.clusters = None

        self.gs_widget.clear()
        self.gs_widget.clear_gene_sets()
        self.cluster_info_view.setModel(None)

        self.cluster_indicators = []
        self.cluster_var = None
        self.batch_indicator = None
        self.cluster_indicator_model.set_domain(None)
        self.batch_indicator_model.set_domain(None)

        self.__update_info_box()

        if data:
            self.input_data = data

            self.cluster_indicator_model.set_domain(self.input_data.domain)
            self.batch_indicator_model.set_domain(self.input_data.domain)

            # For Cluster Indicator do not use categorical variables that contain only one value.
            self.cluster_indicator_model.wrap([
                item for item in self.cluster_indicator_model
                if len(item.values) > 1
            ])
            # First value in batch indicator model is a NoneType,
            # we can skip it when we validate categorical variables
            self.batch_indicator_model.wrap(self.batch_indicator_model[:1] + [
                item for item in self.batch_indicator_model[1:]
                if len(item.values) > 1
            ])

            self.tax_id = self.input_data.attributes.get(TAX_ID, None)
            self.use_attr_names = self.input_data.attributes.get(
                GENE_AS_ATTRIBUTE_NAME, None)
            self.gene_id_attribute = self.input_data.attributes.get(
                GENE_ID_ATTRIBUTE, None)

            if not self.cluster_indicator_model:
                self.Error.no_cluster_indicator()
                return
            elif not self.use_attr_names:
                self.Error.gene_as_attributes()
                return

            self.openContext(self.input_data.domain)

            self.gs_widget.load_gene_sets(self.tax_id)
            if self.cluster_indicator_model and len(
                    self.cluster_indicators) < 1:
                self.cluster_indicators = [self.cluster_indicator_model[0]]
            if self.batch_indicator_model and self.batch_indicator is None:
                self.batch_indicator = self.batch_indicator_model[0]

            self.invalidate()

            if self.custom_data:
                self.refresh_custom_gene_sets()
                self._handle_future_model()
                self.handle_custom_gene_sets()

    @Inputs.custom_sets
    def handle_custom_input(self, data):
        self.Error.clear()
        self.Warning.clear()
        self.closeContext()
        self.custom_data = None
        self.custom_tax_id = None
        self.custom_use_attr_names = None
        self.custom_gene_id_attribute = None
        self.custom_gene_id_column = None
        self.num_of_custom_sets = None
        self.feature_model.set_domain(None)

        if data:
            self.custom_data = data
            self.feature_model.set_domain(self.custom_data.domain)
            self.custom_tax_id = str(
                self.custom_data.attributes.get(TAX_ID, None))
            self.custom_use_attr_names = self.custom_data.attributes.get(
                GENE_AS_ATTRIBUTE_NAME, None)
            self.custom_gene_id_attribute = self.custom_data.attributes.get(
                GENE_ID_ATTRIBUTE, None)
            self.custom_gene_id_column = self.custom_data.attributes.get(
                GENE_ID_COLUMN, None)

            self._handle_future_model()

        if self.input_data:
            self.openContext(self.input_data.domain)

        self.gs_label_combobox.setDisabled(True)
        self.refresh_custom_gene_sets()
        self.handle_custom_gene_sets(select_customs_flag=True)

    def __check_organism_mismatch(self):
        """ Check if organisms from different inputs match.

        :return: True if there is a mismatch
        """
        if self.tax_id is not None and self.custom_tax_id is not None:
            return self.tax_id != self.custom_tax_id
        return False

    def _handle_future_model(self):
        if self.custom_gene_set_indicator in self.feature_model:
            index = self.feature_model.indexOf(self.custom_gene_set_indicator)
            self.custom_gene_set_indicator = self.feature_model[index]
        else:
            if self.feature_model:
                self.custom_gene_set_indicator = self.feature_model[0]
            else:
                self.custom_gene_set_indicator = None

    def handle_custom_gene_sets(self, select_customs_flag=False):
        if self.custom_gene_set_indicator:
            if self.custom_data is not None and self.custom_gene_id_column is not None:

                if self.__check_organism_mismatch():
                    self.gs_label_combobox.setDisabled(True)
                    self.Error.organism_mismatch()
                    self.gs_widget.update_gs_hierarchy()
                    self.__gene_sets_enrichment()
                    return

                if isinstance(self.custom_gene_set_indicator,
                              DiscreteVariable):
                    labels = self.custom_gene_set_indicator.values
                    gene_sets_names = [
                        labels[int(idx)]
                        for idx in self.custom_data.get_column_view(
                            self.custom_gene_set_indicator)[0]
                    ]
                else:
                    gene_sets_names, _ = self.custom_data.get_column_view(
                        self.custom_gene_set_indicator)

                self.num_of_custom_sets = len(set(gene_sets_names))
                gene_names, _ = self.custom_data.get_column_view(
                    self.custom_gene_id_column)
                hierarchy_title = (self.custom_data.name if
                                   self.custom_data.name else 'Custom sets', )
                try:
                    self.gs_widget.add_custom_sets(
                        gene_sets_names,
                        gene_names,
                        hierarchy_title=hierarchy_title,
                        select_customs_flag=select_customs_flag)
                except GeneSetException:
                    pass
                self.gs_label_combobox.setDisabled(False)
            else:
                self.gs_widget.update_gs_hierarchy()

        self.__gene_sets_enrichment()
        self.__update_info_box()

    def refresh_custom_gene_sets(self):
        self.gs_widget.clear_custom_sets()
        # self.gs_widget.update_gs_hierarchy()

    def gene_scores_output(self, selected_clusters):

        metas = [
            StringVariable('Gene'),
            StringVariable(NCBI_ID),
            StringVariable('Rank'),
            ContinuousVariable('Statistic score'),
            ContinuousVariable('P-value'),
            ContinuousVariable('FDR')
        ]

        if len(self.new_cluster_profile):
            # note: order is important
            metas = self.cluster_indicators + metas

        domain = Domain([], metas=metas, class_vars=self.cluster_var)

        data = []
        for cluster in selected_clusters:
            num_of_genes = len(cluster.filtered_genes)

            scores = [gene.score for gene in cluster.filtered_genes]
            p_vals = [gene.p_val for gene in cluster.filtered_genes]
            fdr_vals = [gene.fdr for gene in cluster.filtered_genes]
            gene_names = [gene.input_name for gene in cluster.filtered_genes]
            gene_ids = [gene.ncbi_id for gene in cluster.filtered_genes]
            rank = rankdata(p_vals, method='min')

            if len(self.new_cluster_profile):
                profiles = [[cluster.index] * num_of_genes]
                [
                    profiles.append([p] * num_of_genes)
                    for p in self.new_cluster_profile[cluster.index]
                ]
            else:
                profiles = [[cluster.index] * num_of_genes]

            for row in zip(*profiles, gene_names, gene_ids, rank, scores,
                           p_vals, fdr_vals):
                data.append(list(row))

        out_data = Table(domain, data)
        out_data.attributes[TAX_ID] = self.tax_id
        out_data.attributes[GENE_AS_ATTRIBUTE_NAME] = False
        out_data.attributes[GENE_ID_COLUMN] = NCBI_ID
        self.Outputs.gene_scores.send(out_data)

    def gene_set_scores_output(self, selected_clusters):

        metas = [
            StringVariable('Term'),
            StringVariable('Term ID'),
            StringVariable('Rank'),
            ContinuousVariable('P-value'),
            ContinuousVariable('FDR')
        ]

        if len(self.new_cluster_profile):
            # note: order is important
            metas = self.cluster_indicators + metas

        domain = Domain([], metas=metas, class_vars=self.cluster_var)

        data = []
        for cluster in selected_clusters:
            num_of_sets = len(cluster.filtered_gene_sets)

            p_vals = [gs.p_val for gs in cluster.filtered_gene_sets]
            fdr_vals = [gs.fdr for gs in cluster.filtered_gene_sets]
            gs_names = [gs.name for gs in cluster.filtered_gene_sets]
            gs_ids = [gs.gs_id for gs in cluster.filtered_gene_sets]
            rank = rankdata(p_vals, method='min')

            if len(self.new_cluster_profile):
                profiles = [[cluster.index] * num_of_sets]
                [
                    profiles.append([p] * num_of_sets)
                    for p in self.new_cluster_profile[cluster.index]
                ]
            else:
                profiles = [[cluster.index] * num_of_sets]

            for row in zip(*profiles, gs_names, gs_ids, rank, p_vals,
                           fdr_vals):
                data.append(list(row))

        self.Outputs.gene_set_scores.send(Table(domain, data))

    def commit(self):
        selection_model = self.cluster_info_view.selectionModel()
        selected_rows = selection_model.selectedRows()
        selected_clusters = []
        selected_cluster_indexes = set()
        selected_cluster_genes = set()

        if not self.input_data or not selected_rows:
            self.Outputs.selected_data.send(None)
            return

        for sel_row in selected_rows:
            cluster = sel_row.data()
            selected_clusters.append(cluster)
            selected_cluster_indexes.add(cluster.index)
            [
                selected_cluster_genes.add(gene.ncbi_id)
                for gene in cluster.filtered_genes
            ]

        # get columns of selected clusters
        selected_columns = [
            column for column in self.input_data.domain.attributes
            if self.gene_id_attribute in column.attributes
            and str(column.attributes[
                self.gene_id_attribute]) in selected_cluster_genes
        ]

        domain = Domain(selected_columns, self.input_data.domain.class_vars,
                        self.input_data.domain.metas)
        output_data = self.input_data.from_table(domain, self.input_data)

        # get rows of selected clusters
        selected_rows = [
            row_index
            for row_index, col_index in enumerate(self.rows_by_cluster)
            if col_index in selected_cluster_indexes
        ]

        # send to output signal
        self.Outputs.selected_data.send(output_data[selected_rows])
        self.gene_scores_output(selected_clusters)
        self.gene_set_scores_output(selected_clusters)
    def __init__(self):
        super().__init__()

        # widget attributes
        self.input_data = None
        self.store_input_domain = None
        self.input_genes_names = []
        self.input_genes_ids = []

        self.tax_id = None
        self.use_attr_names = None
        self.gene_id_attribute = None

        # custom gene set input
        self.feature_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, StringVariable))
        self.custom_data = None
        self.custom_tax_id = None
        self.custom_use_attr_names = None
        self.custom_gene_id_attribute = None
        self.custom_gene_id_column = None
        self.num_of_custom_sets = None

        self.rows_by_cluster = None
        self.rows_by_batch = None
        self.clusters = []
        self.new_cluster_profile = []

        # data model
        self.cluster_info_model = None

        # Info
        info_box = vBox(self.controlArea, 'Info')
        self.input_info = widgetLabel(info_box)

        # Cluster selection
        self.cluster_indicator_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, ), separators=False)
        self.cluster_indicator_box = widgetBox(self.controlArea,
                                               'Cluster Indicator')

        self.cluster_indicator_view = listView(
            self.cluster_indicator_box,
            self,
            'cluster_indicators',
            model=self.cluster_indicator_model,
            selectionMode=QListWidget.MultiSelection,
            callback=self.invalidate,
            sizeHint=QSize(256, 70))

        # Batch selection
        self.batch_indicator_model = itemmodels.DomainModel(
            valid_types=(DiscreteVariable, ), separators=False, placeholder="")
        box = widgetBox(self.controlArea, 'Batch Indicator')
        self.batch_indicator_combobox = comboBox(
            box,
            self,
            'batch_indicator',
            model=self.batch_indicator_model,
            sendSelectedValue=True,
            callback=self.batch_indicator_changed)

        # Gene scoring
        box = widgetBox(self.controlArea, 'Gene Scoring')
        self.gene_scoring = GeneScoringWidget(box, self)
        self.gene_scoring.set_method_selection_area('scoring_method_selection')
        self.gene_scoring.set_method_design_area('scoring_method_design')
        self.gene_scoring.set_test_type('scoring_test_type')

        # Gene Sets widget
        gene_sets_box = widgetBox(self.controlArea, "Gene Sets")
        self.gs_widget = GeneSetsSelection(gene_sets_box, self,
                                           'stored_gene_sets_selection')
        self.gs_widget.hierarchy_tree_widget.itemClicked.connect(
            self.__gene_sets_enrichment)

        # custom gene sets area
        box = vBox(self.controlArea, "Custom Gene Sets")

        if self.custom_gene_set_indicator not in self.feature_model:
            self.custom_gene_set_indicator = None

        self.gs_label_combobox = comboBox(
            box,
            self,
            "custom_gene_set_indicator",
            sendSelectedValue=True,
            model=self.feature_model,
            callback=self.handle_custom_gene_sets)
        self.gs_label_combobox.setDisabled(True)

        # main area
        splitter = QSplitter(Qt.Horizontal, self.mainArea)
        self.mainArea.layout().addWidget(splitter)

        genes_filter = widgetBox(splitter,
                                 'Filter Genes',
                                 orientation=QHBoxLayout())
        spin(genes_filter,
             self,
             'max_gene_count',
             0,
             10000,
             label='Count',
             tooltip='Minimum genes count',
             checked='use_gene_count_filter',
             callback=self.filter_genes,
             callbackOnReturn=True,
             checkCallback=self.filter_genes)

        doubleSpin(genes_filter,
                   self,
                   'max_gene_p_value',
                   0.0,
                   1.0,
                   0.0001,
                   label='p-value',
                   tooltip='Maximum p-value of the enrichment score',
                   checked='use_gene_pval_filter',
                   callback=self.filter_genes,
                   callbackOnReturn=True,
                   checkCallback=self.filter_genes)

        doubleSpin(genes_filter,
                   self,
                   'max_gene_fdr',
                   0.0,
                   1.0,
                   0.0001,
                   label='FDR',
                   tooltip='Maximum false discovery rate',
                   checked='use_gene_fdr_filter',
                   callback=self.filter_genes,
                   callbackOnReturn=True,
                   checkCallback=self.filter_genes)

        gene_sets_filter = widgetBox(splitter,
                                     'Filter Gene Sets',
                                     orientation=QHBoxLayout())
        spin(gene_sets_filter,
             self,
             'min_gs_count',
             0,
             DISPLAY_GENE_SETS_COUNT,
             label='Count',
             tooltip='Minimum genes count',
             checked='use_gs_count_filter',
             callback=self.filter_gene_sets,
             callbackOnReturn=True,
             checkCallback=self.filter_gene_sets)

        doubleSpin(gene_sets_filter,
                   self,
                   'max_gs_p_value',
                   0.0,
                   1.0,
                   0.0001,
                   label='p-value',
                   tooltip='Maximum p-value of the enrichment score',
                   checked='use_gs_pval_filter',
                   callback=self.filter_gene_sets,
                   callbackOnReturn=True,
                   checkCallback=self.filter_gene_sets)

        doubleSpin(gene_sets_filter,
                   self,
                   'max_gs_fdr',
                   0.0,
                   1.0,
                   0.0001,
                   label='FDR',
                   tooltip='Maximum false discovery rate',
                   checked='use_gs_max_fdr',
                   callback=self.filter_gene_sets,
                   callbackOnReturn=True,
                   checkCallback=self.filter_gene_sets)

        self.cluster_info_view = QTableView()
        self.cluster_info_view.verticalHeader().setVisible(False)
        self.cluster_info_view.setItemDelegate(HTMLDelegate())
        self.cluster_info_view.horizontalHeader().hide()
        self.cluster_info_view.horizontalHeader().setSectionResizeMode(
            QHeaderView.Stretch)

        auto_commit(self.controlArea,
                    self,
                    "auto_commit",
                    "&Commit",
                    box=False)

        self.mainArea.layout().addWidget(self.cluster_info_view)