The overview by Abi-Dargham et al1 offers a perspective on psychiatric biomarker development primarily based on the classical forward translational model. This approach, currently prioritized by the US National Institute of Mental Health, postulates a linear path from first identifying a biological pathological process through to development of an intervention that both engages the pathophysiologic target and leads to clinical improvement. This model reflects the beauty of translational neuroscience in its most pure form. When successful, it brings great intellectual satisfaction and has the added benefit of enabling a clear explanation for clinicians to use to bolster their treatment recommendations to patients, thereby building confidence in the intervention for all involved. The overview outlines several novel areas of biomarker investigation for mood disorders, though all examples still stand far from the essential third (external validation) and fourth (clinical utility) stages of development. From our vantage point, there are three key clinical questions to prioritize in biomarker development for major depressive disorder (MDD). First, the greatest clinical utility that biomarkers can offer would be to match individuals to the treatment most likely to be efficacious, without harmful effects, commonly characterized as “precision medicine”. Precision medicine is driven by the variability of therapeutic outcomes within the context of a specific diagnosis. There is little controversy around this goal, yet the bulk of biomarkers for MDD to date have focused on distinguishing patients with MDD from healthy controls. This distinction has little clinical relevance, given that non-depressed people do not present to the clinic for care. Second, the primary diagnostic challenge in assessing a patient who presents with a depressed mood is the differential diagnosis of MDD versus bipolar disorder, not versus “non-depressed”. A biomarker that could distinguish these two mood disorders would have high clinical utility, as it would directly inform treatment choices that differ and carry different levels of risk. This differential diagnosis is of greatest importance for adolescents and emerging adults, for whom a major depressive episode may be the presenting mood complaint, yet the life history is too short to have experienced hypomania or mania, and for whom the initiation of an antidepressant medication in the absence of a mood stabilizer when needed could result in tragic consequences. Finally, prognostic biomarkers of MDD course would have great clinical utility for care planning and as targets for intervention. Despite clear clinical evidence of a subset of patients with MDD who have a progressive, deteriorating course of illness that requires treatment with maintenance electroconvulsive therapy or deep brain stimulation, the field has suffered from a severe under-investment in biologically-informed longitudinal studies of MDD that might identify biomarkers of poorer prognosis and increasing treatment resistance. Furthermore, the majority of mood disorder patients being treated today receive antidepressant medications for much longer durations than in the past, but the potentially adverse impacts on disease course of such open-ended neurochemical modulation is almost entirely unknown. Development of biomarkers that indicate probable MDD recurrence, supplemented by behavioral measures, could inform decision-making regarding maintenance or tapering of treatment, and guide the development of interventions targeting the recurrence-risk biology. As the primary organ of interest in mood disorders, characterizing the state of the brain as a component of biomarker development cannot be overstated. Blood-based inflammatory and metabolomic markers have been demonstrated to modulate core neurocircuits in MDD2, 3. Characterizing anatomical and functional neurocircuit states within studies of treatment-selection biomarker development, and linking them to clinical features, may also inform additional strategic reverse translational mechanistic studies, enhancing the likelihood for identifying impactful targets for investigation4-6. Given their high potential clinical utility, how should the field approach developing treatment-selection biomarkers for patients with MDD? The classical approach of identifying a pathological target and engineering agents to engage that target is the one that offers the greatest long-term pay-off, as it brings mental disorders into line with medical disorders for which the pathophysiology is more clearly defined. But, should this be the only way forward for identifying treatment-relevant biomarkers7? The overview by Abi-Dargham et al suggests that the promised land for the application of forward-translation biomarkers is very far off into the future. What can be done today that could yield more rapidly applicable biomarkers with high clinical utility? We believe that there is enormous biomarker discovery value that can derive from the highly divergent mechanisms of action of the existing MDD treatments and the heterogeneity of patient responses to those treatments. Patients may do poorly with psychotherapy and wonderfully with a selective serotonin reuptake inhibitor, and vice versa. Previous studies have already demonstrated that such outcomes reflect not two different brain networks but different states of the same network4, 5, 8, suggesting that classical forward translation approaches alone may prove inadequate. Further, patients showing no response to one class of medication may achieve remission with switch to an alternative class or after addition of an atypical antipsychotic or lithium, while others only improve after receiving a course of ketamine or a neuromodulatory approach. Mechanistic changes that lead to improvement with each of these treatments likely differ, even if, as hypothesized, they all ultimately lead to a final common pathway of enhanced synaptogenesis or other modes of neuroplasticity at whichever depression “brain state” they are introduced. Therefore, critical to the ultimate selection of the “right” psychotherapy, pharmacotherapy or neuromodulation therapy is a deeper understanding and characterization of a patient's current brain state and what renders a patient unable to respond to a treatment or to lose response to a treatment that was previously effective. Differential responses to invasive and non-invasive forms of neuromodulation have particular explanatory power. In fact, in addition to identifying brain states of response to the different forms of stimulation, activity within networks contributing to those states can be directly modulated by the intervention, thus informing a mechanistic forward translational approach to biomarker development9. This rich mix of a variety of effective treatments with differing biological effects, combined with the individual variability of response to specific treatments, offers great value for revealing biomarkers with high clinical utility. Yet, investment in such approaches has been minimal. We suspect that this under-investment is due to the absence of mechanistic hypothesis testing in exploration-based, association-driven methods. Applying a reverse engineering approach, working backwards from treatment outcome heterogeneity to identify treatment-selection biomarkers (which may not have any direct causal role in the pathophysiology of disease), has already proven to be valid and potentially productive4-8. The clinical utility of such biomarkers rests on their potential to allow practitioners to move beyond the current trial-and-error standard, thus shortening time to remission and minimizing exposure to potential adverse effects. Application of treatment-selection biomarkers would further transcend the concept of “clinical stages” of treatment resistance, enabling patients to proceed immediately to more intensive treatments that, under current care models, are withheld until the patient demonstrates non-response to standard treatments. Researchers should remain mindful of the seductive nature of the classical forward translational medical models that proceed linearly from a defined pathophysiology to a treatment. There is truth in their beauty, but the day of their emergence for mood disorders remains distant. In the meantime, there are other, perhaps less intellectually satisfying, approaches that may nevertheless lead to significant gains in treatment selection and mitigation of disease course. Explaining variability in clinical outcomes via biomarkers integrated within well-conducted comparative treatment trials has the greatest immediate potential to inform decision making within a precision medicine framework. Neglecting to invest in and recognize the value of these more imminently impactful, clinically actionable biomarker strategies is a disservice to the millions of patients who are suffering now and will continue to suffer into the future.