ctheodoris

/

Geneformer

like 210

I am trying to reproduce some of your results using the fine-tuned CM cell classifier model in conjugation with the human_dcm_hcm_nf dataset. I tried to delete ADCY5, SRPK3 and ACTB (independently of each other), all of which gave statistically insignificant results. Any guidance would be greatly appreciated.

    isp = InSilicoPerturber(
        perturb_type="delete",
        perturb_rank_shift=None,
        genes_to_perturb=["ENSG00000184343"],
        combos=0,
        anchor_gene=None,
        model_type="CellClassifier",
        num_classes=3,
        emb_mode="cell",
        cell_emb_style="mean_pool",
        filter_data={"cell_type":["Cardiomyocyte1","Cardiomyocyte2","Cardiomyocyte3"]},
        cell_states_to_model={
            'state_key': 'disease', 
            'start_state': 'dcm', 
            'goal_state': 'nf', 
            'alt_states': ['hcm']
        },
        max_ncells=None,
        emb_layer=0,
        forward_batch_size=50,
        nproc=16,
    )

    isp.perturb_data(
        "./CellClassifier_cardiomyopathies/",
        "./human_dcm_hcm_nf.dataset",
        "./perturb_out/",
         prefix
    )

ispstats = InSilicoPerturberStats(mode="goal_state_shift",
                                  genes_perturbed=["ENSG00000184343"],
                                  combos=0,
                                  anchor_gene=None,
                                  cell_states_to_model={"disease":(["dcm"],["nf"],["hcm"])})

ispstats.get_stats("./perturb_out/",
                   None,
                   "./perturb_out/",
                   prefix)

Results produced the cell embedding shifts for each cell expressing the respective gene. Here is a list of the mean values for the Shift_to_goal_end column ± std dev:
ACTB: 0.0013 ± 0.0160
ADCY5: -0.0011 ± 0.0340
SRPK3: -0.00105 ± 0.0246